agentskit.io

# AgentsKit.js

> Foundation library for JavaScript agents. Small packages, one contract, everything composes — UI bindings (7 frameworks), autonomous runtimes, tools, skills, memory, RAG, observability, evaluation, sandboxing. Product chat UI is AgentsKit Chat (sibling), not this monorepo alone.

Six stable contracts (Adapter, Tool, Skill, Memory, Retriever, Runtime). 22 packages under `@agentskit/*`. Install what you need. Zero-dep foundation under the 10 KB gzip budget.

## For agents

If you are an LLM reading this site: start at `/docs/for-agents`. That tab is dense, cross-linked, and designed to fit in one context window per package.

## AgentsKit ecosystem

- [AgentsKit](https://www.agentskit.io) **(current)** — Build agents without gluing many libraries together. Role: `foundation`. Maturity: beta. Machine index: https://www.agentskit.io/llms.txt
- [AgentsKit Registry](https://registry.agentskit.io) — Copy ready-made agents and own the source. Role: `starting-point`. Maturity: beta. Machine index: https://registry.agentskit.io/llms.txt
- [AgentsKit Chat](https://chat.agentskit.io) — Define one agent experience and deliver it across interfaces. Role: `experience`. Maturity: alpha. Machine index: https://chat.agentskit.io/llms.txt
- [Agents Playbook](https://playbook.agentskit.io) — Train coding agents to ship code your team would merge. Role: `discipline`. Maturity: stable. Machine index: https://playbook.agentskit.io/llms.txt
- [Doc Bridge](https://doc-bridge.agentskit.io/) — Turn repository documentation into executable agent handoffs. Role: `understanding`. Maturity: stable. Machine index: https://doc-bridge.agentskit.io/llms.txt
- [AgentsKit Code Review](https://github.com/AgentsKit-io/code-review-cli) — Run deep, low-noise review with the model already in use. Role: `verification`. Maturity: alpha. Machine index: https://raw.githubusercontent.com/AgentsKit-io/code-review-cli/main/llms.txt
- [AgentsKit OS](https://akos.agentskit.io) — Run and govern agents in production. Role: `operation`. Maturity: stable. Machine index: https://akos.agentskit.io/llms.txt

## Get started

- [Architecture at a glance](https://www.agentskit.io/docs/get-started/architecture-at-a-glance): See the AgentsKit ecosystem by layer, how the contracts compose, and how teams usually adopt it over time.
- [AgentsKit vs. alternatives](https://www.agentskit.io/docs/get-started/comparison): How AgentsKit compares to LangChain.js, Vercel AI SDK, Mastra, and LlamaIndex.js across the surface area that actually matters in production.
- [Which package do I need?](https://www.agentskit.io/docs/get-started/decision-tree): A 5-question decision tree from your use case to the one or two AgentsKit packages you actually need to install.
- [@agentskit/core hits 1.0](https://www.agentskit.io/docs/get-started/announcements/core-v1): The JavaScript agent foundation library that doesn't lock you in. Measured ~5.17 KB gzipped at v1 (budget ≤10 KB). Zero dependencies. Six contracts pinned to ADRs.
- [Adapter](https://www.agentskit.io/docs/get-started/concepts/adapter): The seam between AgentsKit and an LLM provider. The same interface for OpenAI, Anthropic, Gemini, Ollama, or anything that streams tokens.
- [Error Handling](https://www.agentskit.io/docs/get-started/concepts/errors): AgentsKit's didactic error system — Rust-compiler-style errors with codes, hints, and doc links baked in.
- [Implementations index](https://www.agentskit.io/docs/get-started/concepts/implementations): Every concrete implementation of every AgentsKit contract. One page to answer "what ships with this?".
- [Concepts](https://www.agentskit.io/docs/get-started/concepts): Six contracts. Every AgentsKit package is an implementation of one of them.
- [Memory](https://www.agentskit.io/docs/get-started/concepts/memory): Two contracts, not one. ChatMemory persists conversation history. VectorMemory stores embeddings for semantic recall. Different problems, different backends.
- [Mental model](https://www.agentskit.io/docs/get-started/concepts/mental-model): The six concepts every AgentsKit user should know.
- [Retriever](https://www.agentskit.io/docs/get-started/concepts/retriever): One narrow interface that serves RAG, BM25, web search, code search, and memory recall through composition.
- [Runtime](https://www.agentskit.io/docs/get-started/concepts/runtime): The conductor. Owns the loop, tool execution, memory persistence, retrieval, delegation, observability, and abort.
- [Skill](https://www.agentskit.io/docs/get-started/concepts/skill): A persona the model becomes — a system prompt plus optional examples, tools, and delegates. Pure declarative configuration.
- [Tool](https://www.agentskit.io/docs/get-started/concepts/tool): A function the model can call, with JSON-schema-typed arguments and a JSON-serializable result.
- [Build your first agent](https://www.agentskit.io/docs/get-started/getting-started/build-your-first-agent): Run a deterministic AgentsKit agent locally, then connect any supported model provider without changing the runtime.
- [For AI Agents](https://www.agentskit.io/docs/get-started/getting-started/for-ai-agents): The entire AgentsKit API in one page. Paste this into your LLM context.
- [Installation](https://www.agentskit.io/docs/get-started/getting-started/installation): Install the AgentsKit packages you need — and only those.
- [Quick start](https://www.agentskit.io/docs/get-started/getting-started/quickstart): Headless useChat binding hello-world in under 10 lines.
- [Start here](https://www.agentskit.io/docs/get-started/getting-started/read-this-first): Three steps to land. Build a real agent, learn the stack, then pick your path.
- [From LangChain.js](https://www.agentskit.io/docs/get-started/migrating/from-langchain): Side-by-side migration guide. Map your LangChain.js code to AgentsKit — with honest callouts about where LangChain still fits.
- [From LangGraph](https://www.agentskit.io/docs/get-started/migrating/from-langgraph): Side-by-side migration guide. Map LangGraph stateful graphs to AgentsKit runtime + topologies.
- [From LlamaIndex](https://www.agentskit.io/docs/get-started/migrating/from-llamaindex): Side-by-side migration guide. Map LlamaIndex retrieval and agent abstractions to AgentsKit RAG + runtime.
- [From Mastra](https://www.agentskit.io/docs/get-started/migrating/from-mastra): Side-by-side migration guide. Map your Mastra code to AgentsKit — with honest callouts about where Mastra still wins.
- [From OpenAI Assistants](https://www.agentskit.io/docs/get-started/migrating/from-openai-assistants): Side-by-side migration guide. Map OpenAI's Assistants API (and the deprecation path) to AgentsKit runtime + memory.
- [From Vercel AI SDK](https://www.agentskit.io/docs/get-started/migrating/from-vercel-ai-sdk): Side-by-side migration guide. Map your Vercel AI SDK code to AgentsKit — with honest callouts about where each wins.
- [Migrating to AgentsKit](https://www.agentskit.io/docs/get-started/migrating): Guides for moving from other frameworks. Honest about what transfers easily and what doesn't.

## Use cases

- [Code agent](https://www.agentskit.io/docs/use-cases/code-agent): Build a coding or code-review agent with filesystem tools, repository context, multi-step runtime behavior, and human oversight.
- [Use cases](https://www.agentskit.io/docs/use-cases): Start from the outcome you want, then see how AgentsKit composes the stack behind it.
- [Internal copilot](https://www.agentskit.io/docs/use-cases/internal-copilot): Build an internal assistant over company data with RAG, permissions, observability, and a path to production.
- [RAG / document Q&A](https://www.agentskit.io/docs/use-cases/rag-app): Chat interface grounded in your own documents — PDFs, Markdown files, knowledge bases.
- [Research agent](https://www.agentskit.io/docs/use-cases/research-agent): Build a research agent that searches, cites, summarizes, and runs as a repeatable job or interactive assistant.
- [Support agent](https://www.agentskit.io/docs/use-cases/support-agent): Build a customer support agent with chat UI, tools, memory, escalation, and production guardrails.

## Compare

- [AgentsKit vs LangChain.js, Vercel AI SDK & assistant-ui](https://www.agentskit.io/docs/compare): An honest comparison of AgentsKit with LangChain.js, the Vercel AI SDK, assistant-ui, and Mastra — bundle size, runtime, contracts, lock-in, and when to pick each.

## UI

- [Angular](https://www.agentskit.io/docs/ui/angular): @agentskit/angular AgentskitChat service — Signals + RxJS BehaviorSubject. Same chat contract as the React useChat hook.
- [ChatContainer](https://www.agentskit.io/docs/ui/chat-container): Headless scrollable transcript container. Auto-scroll on new messages, virtualized-ready.
- [CodeBlock](https://www.agentskit.io/docs/ui/code-block): Headless code block with optional copy button. Style via data-ak-code-block.
- [Data attributes](https://www.agentskit.io/docs/ui/data-attributes): AgentsKit UI components are headless. Every stylable hook is a data-ak-* attribute so your CSS (or your LLM) can target them reliably.
- [InputBar](https://www.agentskit.io/docs/ui/input-bar): Text input + submit. Disabled while streaming. Keyboard shortcuts built in.
- [Markdown](https://www.agentskit.io/docs/ui/markdown): Headless Markdown renderer. Streaming-aware. Style via data-ak-markdown.
- [Message](https://www.agentskit.io/docs/ui/message): Renders one message — user, assistant, tool, or system. Streaming-aware, role-aware.
- [Theming](https://www.agentskit.io/docs/ui/theming): CSS variables + optional preset themes. One source of truth across every framework binding.
- [ThinkingIndicator](https://www.agentskit.io/docs/ui/thinking-indicator): Visibility flag for streaming / reasoning states.
- [ToolCallView](https://www.agentskit.io/docs/ui/tool-call-view): Render a tool invocation — name, args, result, status. Works for all tools.
- [ToolConfirmation](https://www.agentskit.io/docs/ui/tool-confirmation): Human-in-the-loop approval UI for guarded tool calls.
- [useChat](https://www.agentskit.io/docs/ui/use-chat): The one hook every framework binding exposes. Same input, same return, same events.

## Agents

- [Background agents](https://www.agentskit.io/docs/agents/background): Trigger runs on a schedule or HTTP webhook.
- [Multi-Agent Delegation](https://www.agentskit.io/docs/agents/delegation): Coordinate multiple specialist agents from a parent agent using directed delegation.
- [Durable execution](https://www.agentskit.io/docs/agents/durable): Persist every step. Resume after crash. Replay deterministically.
- [Visual flows](https://www.agentskit.io/docs/agents/flow): Author DAGs as YAML, compile to a durable runner. `agentskit flow` is the visual editor + CLI.
- [Guarantees (validator + quota)](https://www.agentskit.io/docs/agents/guarantees): Runtime safety primitives — validator guard wraps regenerable output with a validator chain; per-tool quotas hard-cap blast radius.
- [Human-in-the-loop](https://www.agentskit.io/docs/agents/hitl): Gate tool calls behind approval. Surface reviews to humans before execution.
- [Runtime](https://www.agentskit.io/docs/agents/runtime): Execution engine for autonomous agents — runs a ReAct loop (observe, think, act) until final answer or step limit.
- [Self-debug](https://www.agentskit.io/docs/agents/self-debug): Agents that read their own traces, diagnose failures, retry with corrections.
- [Speculate](https://www.agentskit.io/docs/agents/speculate): Run N candidates in parallel. Pick the best by a user-defined scorer.
- [Topologies](https://www.agentskit.io/docs/agents/topologies): Four proven multi-agent patterns — supervisor, swarm, hierarchical, blackboard.
- [Authoring skills](https://www.agentskit.io/docs/agents/skills/authoring): A skill = prompt + behavior + metadata. Versioned, composable, shippable.
- [clinicalNoteSummarizer](https://www.agentskit.io/docs/agents/skills/clinical-note-summarizer): SOAP-format clinical-note summarizer for clinicians. Preserves verbatim numerics, strips identifiers, never interprets.
- [codeReviewer](https://www.agentskit.io/docs/agents/skills/code-reviewer): Reviews diffs for bugs, security, style. Comments in PR-review format.
- [coder](https://www.agentskit.io/docs/agents/skills/coder): Implements features from specs — TypeScript-first, TDD-leaning, opinionated about code quality.
- [contractReviewer](https://www.agentskit.io/docs/agents/skills/contract-reviewer): Reviews contract drafts for missing clauses, risk flags, and unclear terms. Always defers final sign-off to a licensed attorney.
- [critic](https://www.agentskit.io/docs/agents/skills/critic): Stress-tests proposals. Finds flaws, missing cases, weak assumptions — before they ship.
- [curriculumDesigner](https://www.agentskit.io/docs/agents/skills/curriculum-designer): Designs lesson plans and assessment rubrics for a topic at a target grade level. Bloom-taxonomy aware, accessibility-aware.
- [customerSupport](https://www.agentskit.io/docs/agents/skills/customer-support): First-line customer support — calm, direct, useful. Diagnoses, resolves, or escalates.
- [dataAnalyst](https://www.agentskit.io/docs/agents/skills/data-analyst): Analyzes tabular data — loads, inspects, profiles, and explains findings in plain English.
- [financialAdvisor](https://www.agentskit.io/docs/agents/skills/financial-advisor): Financial-literacy assistant. Explains concepts and trade-offs; refuses tickers, allocations, and 'should you' statements.
- [healthcareAssistant](https://www.agentskit.io/docs/agents/skills/healthcare-assistant): Information-only patient-facing assistant. Refuses diagnosis, dosage, emergency triage.
- [Skills](https://www.agentskit.io/docs/agents/skills): Personas as packages — system prompt + behavior, versioned and composable.
- [legalAssistant](https://www.agentskit.io/docs/agents/skills/legal-assistant): Information-only legal assistant. Refuses jurisdiction-specific advice; always recommends a licensed attorney for binding decisions.
- [listingConcierge](https://www.agentskit.io/docs/agents/skills/listing-concierge): Helps buyers / renters narrow listings by criteria, schedule tours, and request more info. Hard fair-housing rails.
- [marketAnalyst](https://www.agentskit.io/docs/agents/skills/market-analyst): Pulls comps, computes price-per-sqft, and produces buyer / seller market briefs for a named area. Describes only — no predictions.
- [Skill marketplace](https://www.agentskit.io/docs/agents/skills/marketplace): Publish, install, version skills. Semver range resolution.
- [merchandisingAnalyst](https://www.agentskit.io/docs/agents/skills/merchandising-analyst): Analyses sales / inventory data to surface restock priorities, slow-moving SKUs, and bundle opportunities. Outputs CSV-friendly tables.
- [Ready-made personas](https://www.agentskit.io/docs/agents/skills/personas): Nine skills bundled with @agentskit/skills. Importable, composable.
- [planner](https://www.agentskit.io/docs/agents/skills/planner): Decomposes vague goals into ordered steps with clear success criteria. Ideal as the top node in a supervisor topology.
- [prReviewer](https://www.agentskit.io/docs/agents/skills/pr-reviewer): Reviews a diff against the AgentsKit Manifesto + package CONVENTIONS. Flags violations and suggests concrete rewrites.
- [researcher](https://www.agentskit.io/docs/agents/skills/researcher): Methodical web-search persona that finds, cross-references, and summarizes with citations.
- [securityAuditor](https://www.agentskit.io/docs/agents/skills/security-auditor): Security review of code or config — injection, auth, secrets, crypto, SSRF, supply chain, LLM-specific risks.
- [sqlAnalyst](https://www.agentskit.io/docs/agents/skills/sql-analyst): Read-only data analyst — schema discovery, safe SELECTs, plain-English explanations.
- [sqlGen](https://www.agentskit.io/docs/agents/skills/sql-gen): Natural language → SQL with dialect awareness + safety rails.
- [storefrontConcierge](https://www.agentskit.io/docs/agents/skills/storefront-concierge): Customer-facing storefront agent. Recommends products, looks up order status, and escalates returns/refunds to a human.
- [summarizer](https://www.agentskit.io/docs/agents/skills/summarizer): Compresses long inputs into short, faithful summaries. Length-aware; preserves citations.
- [technicalWriter](https://www.agentskit.io/docs/agents/skills/technical-writer): Skim-friendly technical docs — TL;DR-first, no marketing voice, code-example-driven.
- [transactionTriage](https://www.agentskit.io/docs/agents/skills/transaction-triage): Bookkeeping triage — categorizes a single transaction into a chart-of-accounts entry. Refuses payment / refund / chargeback decisions.
- [translator](https://www.agentskit.io/docs/agents/skills/translator): High-quality translation between natural languages. Preserves formatting, tone, terminology. Includes translatorWithGlossary for forced term overrides.
- [tutor](https://www.agentskit.io/docs/agents/skills/tutor): Socratic tutor. Defaults to questions and hints; only gives direct answers when the user explicitly asks for them.
- [Authoring tools](https://www.agentskit.io/docs/agents/tools/authoring): Define, validate, compose, guard. One contract, multiple flavors.
- [Built-in tools](https://www.agentskit.io/docs/agents/tools/builtins): Ship-ready tools — web, fetch, filesystem, shell.
- [Tools](https://www.agentskit.io/docs/agents/tools): Every function the agent can call. Built-ins, integrations, MCP bridge, composers.
- [MCP bridge](https://www.agentskit.io/docs/agents/tools/mcp): Consume or publish Model Context Protocol tools. Interop with Claude Desktop, Cursor, Continue, etc.
- [airtable](https://www.agentskit.io/docs/agents/tools/integrations/airtable): Airtable — list and create records in any table by base id.
- [browserAgent](https://www.agentskit.io/docs/agents/tools/integrations/browser-agent): Puppeteer-backed browser — goto, click, fill, read, screenshot, wait. For agents that need to interact with JS-rendered pages.
- [cloudflareR2](https://www.agentskit.io/docs/agents/tools/integrations/cloudflare-r2): Cloudflare R2 storage — S3-compatible at the protocol level, mirrors the s3 tool surface.
- [coingecko](https://www.agentskit.io/docs/agents/tools/integrations/coingecko): CoinGecko — crypto prices + market charts. Free tier, no key required.
- [confluence](https://www.agentskit.io/docs/agents/tools/integrations/confluence): Confluence pages — CQL search + create. Basic auth (email + API token).
- [deepgram](https://www.agentskit.io/docs/agents/tools/integrations/deepgram): Deepgram STT — low-latency transcription with speaker diarization. Preferred for realtime + voice-agent flows.
- [discord](https://www.agentskit.io/docs/agents/tools/integrations/discord): Discord Bot API — post messages. For community bots, notifier agents, and interactive slash-command workflows.
- [documentParsers](https://www.agentskit.io/docs/agents/tools/integrations/document-parsers): PDF, DOCX, XLSX parsers — BYO parser functions keep core dependency-free.
- [elevenlabs](https://www.agentskit.io/docs/agents/tools/integrations/elevenlabs): ElevenLabs text-to-speech — high-quality voices in 30+ languages. For narration, voice agents, IVR flows.
- [email](https://www.agentskit.io/docs/agents/tools/integrations/email): Provider-agnostic SMTP send + IMAP fetch. BYO transport (nodemailer, imapflow, Resend, SES) — drivers stay out of the bundle.
- [figma](https://www.agentskit.io/docs/agents/tools/integrations/figma): Figma — read file node tree and export node ids as image URLs. Personal access token.
- [firecrawl](https://www.agentskit.io/docs/agents/tools/integrations/firecrawl): Firecrawl — scrape a URL or crawl a site, returning clean markdown. The workhorse for RAG ingestion from the open web.
- [githubActions](https://www.agentskit.io/docs/agents/tools/integrations/github-actions): GitHub Actions — list runs and trigger workflow_dispatch events.
- [github](https://www.agentskit.io/docs/agents/tools/integrations/github): GitHub REST v3 — search issues, create issues, comment. Pairs with HITL for ship-gating bots.
- [gmail](https://www.agentskit.io/docs/agents/tools/integrations/gmail): Gmail API — list messages, send email. For inbox-triage agents + notification flows.
- [googleCalendar](https://www.agentskit.io/docs/agents/tools/integrations/google-calendar): Google Calendar API — list events, create events. For scheduling agents + standup prep bots.
- [hubspot](https://www.agentskit.io/docs/agents/tools/integrations/hubspot): HubSpot CRM — search contacts and create deals. Private app access token.
- [Integrations](https://www.agentskit.io/docs/agents/tools/integrations): 50 verified connectors for the services agents actually need. Each follows the same contract — install, config, execute — and ships granular sub-tools alongside a bundled set.
- [jira](https://www.agentskit.io/docs/agents/tools/integrations/jira): Jira issues — search via JQL, create. Basic auth (email + API token).
- [linearTriage](https://www.agentskit.io/docs/agents/tools/integrations/linear-triage): Linear triage — list a team's triage queue, assign owner / state / priority.
- [linear](https://www.agentskit.io/docs/agents/tools/integrations/linear): Linear GraphQL — search issues, create issues. For triage, sprint-planning, release-notes agents.
- [maps](https://www.agentskit.io/docs/agents/tools/integrations/maps): Geocoding + reverse-geocoding via any provider with an HTTP API (Google, Mapbox, OpenCage, Positionstack, Nominatim).
- [notion](https://www.agentskit.io/docs/agents/tools/integrations/notion): Notion API — search workspace, create pages. Pair with RAG to turn Notion into agent-queryable memory.
- [openaiImages](https://www.agentskit.io/docs/agents/tools/integrations/openai-images): OpenAI Images API (DALL-E / gpt-image) — generate images from prompts.
- [pagerduty](https://www.agentskit.io/docs/agents/tools/integrations/pagerduty): PagerDuty Events API v2 — trigger / acknowledge / resolve. Optional REST oncall lookup.
- [postgres-cdc](https://www.agentskit.io/docs/agents/tools/integrations/postgres-cdc): Postgres change-data-capture — logical replication slots (pgoutput / wal2json) and Supabase realtime, exposed as tool primitives + a stream helper.
- [postgresWithRoles](https://www.agentskit.io/docs/agents/tools/integrations/postgres-roles): Read/write split for the postgres tool — two role-bound clients, two distinct tools, least-privilege at the database level.
- [postgres](https://www.agentskit.io/docs/agents/tools/integrations/postgres): Postgres query tool — BYO runner keeps the adapter client-agnostic (`postgres.js`, `pg`, Drizzle, Prisma, Neon).
- [reader](https://www.agentskit.io/docs/agents/tools/integrations/reader): Jina Reader — turn any URL into clean markdown. Free, fast, no API key required for casual use.
- [s3](https://www.agentskit.io/docs/agents/tools/integrations/s3): AWS S3 — get, put, list objects. BYO client keeps the tool AWS-SDK-version-agnostic.
- [sentry](https://www.agentskit.io/docs/agents/tools/integrations/sentry): Sentry — search org / project issues and resolve them. Bearer auth-token.
- [shopify](https://www.agentskit.io/docs/agents/tools/integrations/shopify): Shopify Admin — search products and list orders. Custom app access token.
- [slack](https://www.agentskit.io/docs/agents/tools/integrations/slack): Slack Web API — post messages, search history. For notifier bots + conversational search over a workspace.
- [stripeWebhook](https://www.agentskit.io/docs/agents/tools/integrations/stripe-webhook): Verify Stripe webhook signatures and parse the event before the agent ever sees it.
- [stripe](https://www.agentskit.io/docs/agents/tools/integrations/stripe): Stripe API — create customers, payment intents. For checkout assistants + dunning agents.
- [teams](https://www.agentskit.io/docs/agents/tools/integrations/teams): Microsoft Teams — Incoming Webhook (one-way) or Bot Framework (bidirectional). MessageCard + Adaptive Card support.
- [twilio](https://www.agentskit.io/docs/agents/tools/integrations/twilio): Twilio SMS — send via the REST Messages endpoint with E.164 validation.
- [weather](https://www.agentskit.io/docs/agents/tools/integrations/weather): Current weather by lat/lng. Free tier via open-meteo by default; swap provider via config.
- [whisper](https://www.agentskit.io/docs/agents/tools/integrations/whisper): OpenAI Whisper — speech-to-text for audio transcription. 99 languages.

## Data

- [anthropic](https://www.agentskit.io/docs/data/providers/anthropic): Anthropic chat adapter — Claude Opus / Sonnet / Haiku. Streaming, tool-calls, vision, long-context.
- [azureOpenAI](https://www.agentskit.io/docs/data/providers/azure-openai): Azure OpenAI — deployment-routed, api-version-pinned. Drop-in for enterprise Azure tenants.
- [bail (Qwen)](https://www.agentskit.io/docs/data/providers/bail): Alibaba Bailian — Qwen-2.5/3 + Qwen-VL via the DashScope OpenAI-compatibility endpoint.
- [bedrock](https://www.agentskit.io/docs/data/providers/bedrock): AWS Bedrock — enterprise path for Claude on Anthropic's hosted-on-AWS surface.
- [cerebras](https://www.agentskit.io/docs/data/providers/cerebras): Cerebras — wafer-scale chips, OpenAI-compatible. Sub-100ms first-token latency on Llama / Qwen.
- [Choosing an adapter](https://www.agentskit.io/docs/data/providers/choosing): Capability decision table and rules of thumb for picking a chat adapter.
- [cohere](https://www.agentskit.io/docs/data/providers/cohere): Cohere Command — enterprise-focused RAG-friendly models with citations.
- [deepseekEmbedder](https://www.agentskit.io/docs/data/providers/deepseek-embedder): DeepSeek embedding model. Compatible with createOpenAICompatibleEmbedder under the hood.
- [deepseek](https://www.agentskit.io/docs/data/providers/deepseek): DeepSeek — cost-efficient reasoning + chat models. OpenAI-compatible API.
- [Embedders](https://www.agentskit.io/docs/data/providers/embedders): Turn text into vectors. Used by RAG + vector memory.
- [fireworks](https://www.agentskit.io/docs/data/providers/fireworks): Fireworks AI — fast open-model inference with fine-tuning + function-calling support.
- [geminiEmbedder](https://www.agentskit.io/docs/data/providers/gemini-embedder): Google Gemini embeddings — text-embedding-004. Strong multilingual.
- [gemini](https://www.agentskit.io/docs/data/providers/gemini): Google Gemini chat adapter — Gemini 2.5 Pro / Flash. Streaming, tool-calls, vision, 1M+ context.
- [generic](https://www.agentskit.io/docs/data/providers/generic): Turn any `ReadableStream` into an AgentsKit adapter — the lowest-level escape hatch.
- [grokEmbedder](https://www.agentskit.io/docs/data/providers/grok-embedder): xAI embedding model.
- [grok](https://www.agentskit.io/docs/data/providers/grok): xAI Grok — streaming chat + tool calls. OpenAI-compatible API surface.
- [groq](https://www.agentskit.io/docs/data/providers/groq): Groq — ultra-low-latency inference on custom LPU hardware. Llama + Mixtral + Gemma.
- [Higher-order adapters](https://www.agentskit.io/docs/data/providers/higher-order): Compose adapters — route, ensemble, fallback.
- [Hosted chat adapters](https://www.agentskit.io/docs/data/providers/hosted): 17 managed-LLM adapters. Same contract; swap by changing one import.
- [huggingface](https://www.agentskit.io/docs/data/providers/huggingface): Hugging Face Inference Endpoints + Serverless — run any HF-hosted chat model.
- [Providers](https://www.agentskit.io/docs/data/providers): 25 native chat and embedder adapters, plus higher-order adapters that compose candidates. Separate from the 140-provider models catalog.
- [kimiEmbedder](https://www.agentskit.io/docs/data/providers/kimi-embedder): Moonshot Kimi embedding model.
- [kimi](https://www.agentskit.io/docs/data/providers/kimi): Moonshot AI Kimi — long-context, OpenAI-compatible.
- [langchain](https://www.agentskit.io/docs/data/providers/langchain): Wrap any LangChain.js `ChatModel` as an AgentsKit adapter.
- [langgraph](https://www.agentskit.io/docs/data/providers/langgraph): Wrap a LangGraph graph as an AgentsKit adapter — streamEvents surfaced as chunks.
- [llamacpp](https://www.agentskit.io/docs/data/providers/llamacpp): llama.cpp server — run GGUF models on CPU or GPU with minimal overhead.
- [lmstudio](https://www.agentskit.io/docs/data/providers/lmstudio): LM Studio — desktop app that exposes an OpenAI-compatible server over local models.
- [Local runtimes](https://www.agentskit.io/docs/data/providers/local): Run fully offline — Ollama, LM Studio, vLLM, llama.cpp.
- [mistral](https://www.agentskit.io/docs/data/providers/mistral): Mistral — open-weights-friendly family including Mistral Large, Codestral, Mixtral.
- [ollamaEmbedder](https://www.agentskit.io/docs/data/providers/ollama-embedder): Ollama-hosted embedding models — nomic-embed-text, mxbai-embed-large, all-minilm.
- [ollama](https://www.agentskit.io/docs/data/providers/ollama): Ollama — local LLMs on your laptop. Zero-cost, offline, private.
- [createOpenAICompatibleEmbedder](https://www.agentskit.io/docs/data/providers/openai-compatible-embedder): Escape hatch for any OpenAI-compatible `/v1/embeddings` endpoint — LMS, vLLM, Cerebras, Voyage, Jina, internal gateways.
- [openaiEmbedder](https://www.agentskit.io/docs/data/providers/openai-embedder): OpenAI text embeddings — text-embedding-3-small / -large. Default choice for most RAG stacks.
- [openai](https://www.agentskit.io/docs/data/providers/openai): OpenAI chat adapter — GPT-4o, o-series, GPT-5. Streaming, tool-calls, parallel tools, multimodal.
- [openrouter](https://www.agentskit.io/docs/data/providers/openrouter): OpenRouter — one API to route across 100+ providers with transparent pricing.
- [replicate](https://www.agentskit.io/docs/data/providers/replicate): Replicate — hosted open models behind one API. Two-step prediction + SSE stream.
- [together](https://www.agentskit.io/docs/data/providers/together): Together AI — hosts 200+ open models behind one OpenAI-compatible API.
- [vercelAI](https://www.agentskit.io/docs/data/providers/vercel-ai): Wrap a Vercel AI SDK `LanguageModel` as an AgentsKit adapter.
- [vertex](https://www.agentskit.io/docs/data/providers/vertex): Vertex AI — Gemini on Google Cloud with project + region routing and OAuth2 tokens.
- [vllm](https://www.agentskit.io/docs/data/providers/vllm): vLLM — high-throughput self-hosted inference with OpenAI-compatible API. For production workloads on your own GPUs.
- [webllm (browser-only / WebGPU)](https://www.agentskit.io/docs/data/providers/webllm): 100% on-device inference via WebLLM (MLC). No API key, no token cost, no inference network call.
- [Chunking](https://www.agentskit.io/docs/data/rag/chunking): Split docs before embedding. Sensible defaults; override per doc type.
- [Context injection](https://www.agentskit.io/docs/data/rag/context-injection): How retrieved chunks become system prompt context — manual pattern and runtime integration.
- [createRAG](https://www.agentskit.io/docs/data/rag/create-rag): One-liner RAG pipeline — chunk, embed, store, retrieve, search.
- [Hybrid search](https://www.agentskit.io/docs/data/rag/hybrid): Vector + BM25 blend with weighted normalization.
- [RAG](https://www.agentskit.io/docs/data/rag): Plug-and-play retrieval. Chunk, embed, search, rerank, loaders.
- [Document loaders](https://www.agentskit.io/docs/data/rag/loaders): Fetch + normalize documents from URLs, GitHub, Notion, Confluence, Google Drive, S3, GCS, Dropbox, OneDrive, PDFs.
- [Rerank](https://www.agentskit.io/docs/data/rag/rerank): Post-retrieval reranker. Cohere / BGE / built-in BM25.
- [createAutoSummarizingMemory](https://www.agentskit.io/docs/data/memory/auto-summarize): Fold oldest messages into a running summary. Token-budget-friendly.
- [chroma](https://www.agentskit.io/docs/data/memory/chroma): Chroma vector DB via HTTP.
- [createEncryptedMemory](https://www.agentskit.io/docs/data/memory/encrypted): AES-GCM-256 envelope over any ChatMemory. Keys never touch disk in plaintext.
- [fileChatMemory](https://www.agentskit.io/docs/data/memory/file-chat): JSON-file-backed chat memory. Zero infra. Survives restarts.
- [fileVectorMemory](https://www.agentskit.io/docs/data/memory/file-vector): Pure-JS file-persisted vector store. Zero infra.
- [createInMemoryGraph](https://www.agentskit.io/docs/data/memory/graph): Knowledge graph memory — nodes + edges, queryable.
- [createHierarchicalMemory](https://www.agentskit.io/docs/data/memory/hierarchical): MemGPT-style tiers — working / recall / archival.
- [Memory](https://www.agentskit.io/docs/data/memory): Chat memory + vector stores + wrappers that make long-running agents practical.
- [LanceDB](https://www.agentskit.io/docs/data/memory/lancedb): LanceDB vector backend for @agentskit/memory — planned via custom adapter.
- [milvusVectorStore](https://www.agentskit.io/docs/data/memory/milvus-vector): Milvus / Zilliz Cloud vector store. v2 REST API with Bearer auth.
- [mongoAtlasVectorStore](https://www.agentskit.io/docs/data/memory/mongo-atlas-vector): MongoDB Atlas Vector Search. Caller injects a Collection-shaped client; \`mongodb\` driver stays as an external concern.
- [createInMemoryPersonalization](https://www.agentskit.io/docs/data/memory/personalization): Per-subject profile store. Inject into system prompt.
- [pgvector](https://www.agentskit.io/docs/data/memory/pgvector): Postgres + pgvector adapter. BYO SQL runner.
- [pinecone](https://www.agentskit.io/docs/data/memory/pinecone): Managed vector DB. Namespaces + metadata filters.
- [qdrant](https://www.agentskit.io/docs/data/memory/qdrant): Self-hosted or cloud Qdrant via HTTP.
- [redisChatMemory](https://www.agentskit.io/docs/data/memory/redis-chat): Redis-backed chat memory for multi-instance + serverless.
- [redisVectorMemory](https://www.agentskit.io/docs/data/memory/redis-vector): Redis Stack / Redis 8+ vector index. Metadata filtering + HNSW.
- [sqliteChatMemory](https://www.agentskit.io/docs/data/memory/sqlite): SQLite-backed chat memory. Indexed by session + timestamp.
- [supabaseVectorStore](https://www.agentskit.io/docs/data/memory/supabase-vector): Supabase-hosted pgvector with direct mutations and a purpose-specific similarity-search RPC.
- [tursoChatMemory](https://www.agentskit.io/docs/data/memory/turso): libSQL / Turso-backed chat memory. Replicated SQLite with the same surface as sqliteChatMemory.
- [upstashVector](https://www.agentskit.io/docs/data/memory/upstash-vector): Serverless HTTP vector DB from Upstash.
- [Metadata filters (VectorFilter v1)](https://www.agentskit.io/docs/data/memory/vector-filters): Normalized metadata-filter shape across every vector backend. One contract, ten translations.
- [createVirtualizedMemory](https://www.agentskit.io/docs/data/memory/virtualized): Hot-window + cold retriever. Keep recent messages; retrieve old ones on demand.
- [weaviateVectorStore](https://www.agentskit.io/docs/data/memory/weaviate-vector): Weaviate vector store. REST batch insert + GraphQL nearVector search.

## Production

- [Edge bundle (Cloudflare Workers / Deno Deploy)](https://www.agentskit.io/docs/production/edge): Ship AgentsKit on Cloudflare Workers, Deno Deploy, Vercel Edge, or Bun. Sub-50KB hot path; tree-shake everything you don't use.
- [VS Code · Raycast · embedded](https://www.agentskit.io/docs/production/embedded): Run AgentsKit agents inside your editor or launcher. Three integration patterns + a worked example for each.
- [Performance budgets](https://www.agentskit.io/docs/production/performance): Bundle size ceilings per package, enforced in CI via size-limit. Measured values injected when available.
- [On-call runbooks](https://www.agentskit.io/docs/production/runbooks): First-response playbooks for the four most common AgentsKit production incidents — LLM provider outage, tool flapping, cost spike, prompt injection.
- [Shipping checklist](https://www.agentskit.io/docs/production/shipping-checklist): A practical checklist for taking an AgentsKit agent from prototype to production.
- [agentskit ai](https://www.agentskit.io/docs/production/cli/ai): Natural language → typed AgentSchema + scaffold.
- [agentskit chat](https://www.agentskit.io/docs/production/cli/chat): Interactive terminal chat. Ink-based. No setup required.
- [agentskit dev](https://www.agentskit.io/docs/production/cli/dev): Local dev server with hot-reload. Devtools + trace feed attached.
- [agentskit doctor](https://www.agentskit.io/docs/production/cli/doctor): Diagnose your environment. Providers, keys, tooling, versions.
- [CLI](https://www.agentskit.io/docs/production/cli): Nine commands wrapping every part of AgentsKit — init, chat, run, dev, doctor, ai, tunnel, rag, config.
- [agentskit init](https://www.agentskit.io/docs/production/cli/init): Scaffold a new project — 9 starter templates spanning React, SvelteKit, Nuxt, Ink, Vite+Ink, Cloudflare Workers, Bun, runtime, and multi-agent.
- [Programmatic API](https://www.agentskit.io/docs/production/cli/programmatic-api): Drive every agentskit subcommand from your own code — same internals as the CLI, exposed as named exports.
- [agentskit run](https://www.agentskit.io/docs/production/cli/run): One-shot agent run from the terminal. Pipe inputs, capture outputs.
- [CI reporters](https://www.agentskit.io/docs/production/evals/ci): Write eval results as JUnit XML, Markdown, or GitHub annotations so failures block PRs and surface in pull request checks.
- [Evals](https://www.agentskit.io/docs/production/evals): Run eval suites against any async agent function, replay recorded sessions in CI, and track prompt regressions with snapshots.
- [Deterministic replay](https://www.agentskit.io/docs/production/evals/replay): Record LLM responses to a cassette file, then replay them in CI without network calls for fast, deterministic tests.
- [Prompt snapshots + diff](https://www.agentskit.io/docs/production/evals/snapshots): Assert that rendered prompts haven't changed unexpectedly, and trace exactly which edit caused a drift.
- [Eval suites](https://www.agentskit.io/docs/production/evals/suites): Define cases with inputs and assertions, then run them against any async agent function to get pass rates and latency metrics.
- [Signed audit log](https://www.agentskit.io/docs/production/observability/audit-log): Hash-chained, HMAC-signed log that makes it cryptographically detectable if any record is altered or deleted.
- [Cost + token accounting](https://www.agentskit.io/docs/production/observability/cost-guard): Set a hard dollar ceiling per run and track token usage with either a zero-dep heuristic or provider-accurate counters.
- [Devtools server](https://www.agentskit.io/docs/production/observability/devtools): Expose a live SSE stream of agent events so any browser-based dashboard can display them in real time.
- [Observability](https://www.agentskit.io/docs/production/observability): Attach loggers, tracers, and cost guards to any runtime — no code changes beyond adding an observer.
- [Loggers + tracers](https://www.agentskit.io/docs/production/observability/loggers): Attach console, LangSmith, or OpenTelemetry observers to any runtime — mix and match, all receive the same event stream.
- [Trace viewer](https://www.agentskit.io/docs/production/observability/trace-viewer): Generate a self-contained HTML trace file from any run — inspect spans offline without a tracing backend.
- [Security](https://www.agentskit.io/docs/production/security): Six primitives for production agents: PII redaction, injection detection, rate limiting, audit log, sandbox enforcement, and HITL approvals.
- [Input validation](https://www.agentskit.io/docs/production/security/input-validation): Opt-in JSON Schema validation of tool inputs plus user-message limits, prompt-injection checks, and allowlists.
- [Mandatory sandbox](https://www.agentskit.io/docs/production/security/mandatory-sandbox): Wrap tools with allow/deny lists and validators so agents cannot run unrestricted side effects.
- [PII redaction](https://www.agentskit.io/docs/production/security/pii-redaction): Strip emails, phones, SSNs, and API keys from messages before they reach the model or get written to logs.
- [Prompt injection](https://www.agentskit.io/docs/production/security/prompt-injection): Detect instruction-hijacking patterns in user input, tool results, and RAG chunks before they reach the model.
- [Rate limiting](https://www.agentskit.io/docs/production/security/rate-limiting): Token-bucket rate limiter keyed by user, IP, or API key — with Redis/Upstash support for multi-host deployments.
- [Sandbox: deep dive](https://www.agentskit.io/docs/production/security/sandbox): How @agentskit/sandbox executes untrusted code — backends, policies, limits, and honest isolation claims.
- [Secret management](https://www.agentskit.io/docs/production/security/secrets): Env var hygiene, vault integrations, rotation, and keeping credentials out of logs and source.
- [SSO (OIDC + SAML)](https://www.agentskit.io/docs/production/security/sso): Verify OIDC ID tokens and SAML assertions to map an inbound request to a tenant. Pure, dependency-free, WebCrypto-only.

## For agents

- [@agentskit/adapters — for agents](https://www.agentskit.io/docs/for-agents/adapters): Provider adapters (OpenAI-compatible + native) + router + ensemble + fallback + generic factory.
- [@agentskit/angular — for agents](https://www.agentskit.io/docs/for-agents/angular): Angular 18+ service exposing chat state as a WritableSignal and RxJS Observable — partial-Ivy AOT/APF, standalone-component friendly.
- [@agentskit/cli — for agents](https://www.agentskit.io/docs/for-agents/cli): agentskit CLI — init, chat, run, dev, doctor, ai, tunnel, rag, config.
- [@agentskit/core — for agents](https://www.agentskit.io/docs/for-agents/core): Zero-dependency foundation. Contracts, chat controller, primitives, and a dozen feature subpaths.
- [@agentskit/eval/braintrust — for agents](https://www.agentskit.io/docs/for-agents/eval-braintrust): Braintrust scoring pipeline — 4 quality + 4 robustness scorers, runner, and CI regression helpers.
- [@agentskit/eval — for agents](https://www.agentskit.io/docs/for-agents/eval): Evaluation harness + deterministic replay + snapshot testing + diff + CI reporters.
- [@agentskit/ink — for agents](https://www.agentskit.io/docs/for-agents/ink): Terminal chat UI via Ink — same `useChat` contract as `@agentskit/react`, running in Node with ANSI theming and keyboard navigation.
- [@agentskit/integrations — for agents](https://www.agentskit.io/docs/for-agents/integrations): Unified single-descriptor service-integration catalog projected into tools, connectors, triggers, and OAuth.
- [@agentskit/mcp — for agents](https://www.agentskit.io/docs/for-agents/mcp): Expose AgentsKit tools as an MCP server over stdio for Claude Desktop, Cursor, Windsurf.
- [@agentskit/memory — for agents](https://www.agentskit.io/docs/for-agents/memory): Chat + vector memory backends, plus hierarchical, encrypted, graph, and personalization stores.
- [@agentskit/observability/langfuse — for agents](https://www.agentskit.io/docs/for-agents/observability-langfuse): Langfuse tracing adapter — plan / tool / model / HITL spans with token, cost, latency capture and parent linking across multi-agent topologies.
- [@agentskit/observability — for agents](https://www.agentskit.io/docs/for-agents/observability): Console + LangSmith + OpenTelemetry logging, token counters, cost guard, trace viewer, signed audit log, devtools server.
- [@agentskit/rag — for agents](https://www.agentskit.io/docs/for-agents/rag): Plug-and-play RAG. Chunking + ingest + retrieve + rerank + hybrid + eleven document loaders.
- [@agentskit/react-native — for agents](https://www.agentskit.io/docs/for-agents/react-native): React Native / Expo binding — same `useChat` contract as `@agentskit/react`, Metro + Hermes safe with streaming polyfill guidance.
- [@agentskit/react — for agents](https://www.agentskit.io/docs/for-agents/react): React hooks + headless chat components driving createChatController.
- [@agentskit/runtime — for agents](https://www.agentskit.io/docs/for-agents/runtime): Standalone agent runtime (ReAct loop) + speculate + topologies + durable execution + background agents.
- [@agentskit/sandbox — for agents](https://www.agentskit.io/docs/for-agents/sandbox): Secure code execution (E2B / Web Worker / local runtimes) + mandatory sandbox policy wrapper.
- [@agentskit/skills — for agents](https://www.agentskit.io/docs/for-agents/skills): Ready-made personas (system prompt + behavior) + composition + marketplace registry.
- [@agentskit/solid — for agents](https://www.agentskit.io/docs/for-agents/solid): SolidJS binding — `useChat` hook backed by createStore for fine-grained reactivity.
- [@agentskit/statechart — for agents](https://www.agentskit.io/docs/for-agents/statechart): Framework-neutral deterministic interaction state and validated snapshots.
- [@agentskit/svelte — for agents](https://www.agentskit.io/docs/for-agents/svelte): Svelte 5 binding — `createChatStore` produces a Readable that drives any Svelte component.
- [@agentskit/templates — for agents](https://www.agentskit.io/docs/for-agents/templates): Authoring toolkit — validated factories + secure on-disk scaffolds for custom tools, skills, adapters, memory, embedders, and flows.
- [@agentskit/tools — for agents](https://www.agentskit.io/docs/for-agents/tools): Built-in executable tools + 20+ third-party integrations + bidirectional MCP bridge.
- [@agentskit/tools/validation — for agents](https://www.agentskit.io/docs/for-agents/validation): Opt-in runtime validation of tool-call arguments against their JSON Schema. Wraps Ajv; plugs into the core ArgsValidator contract (ADR-0008).
- [@agentskit/vue — for agents](https://www.agentskit.io/docs/for-agents/vue): Vue 3 composable — `useChat` backed by `reactive()` with auto-cleanup on scope dispose, plus a headless `<ChatContainer>` component.

## Reference

- [Changelog](https://www.agentskit.io/docs/reference/changelog): Release notes across every @agentskit package, generated from CHANGELOG.md files.
- [Stability levels](https://www.agentskit.io/docs/reference/stability): Every AgentsKit package declares an alpha / beta / stable level. This page defines each level plus the promotion / demotion criteria.
- [Commit style](https://www.agentskit.io/docs/reference/contribute/commit-style): Conventional commits, changesets, and the PR checklist.
- [Docs MDX components](https://www.agentskit.io/docs/reference/contribute/docs-components): Embeddable components available inside every .mdx file — Stackblitz, CodeSandbox, GIFs, Mermaid.
- [Good first issues](https://www.agentskit.io/docs/reference/contribute/good-first-issues): Curated issues ready to grab. Pick one, comment on it to claim, and ship a PR.
- [Translating the docs](https://www.agentskit.io/docs/reference/contribute/i18n): How the AgentsKit docs ship in multiple languages — current status, adding a page, promoting a locale from planned to seed to partial to full.
- [Contribute](https://www.agentskit.io/docs/reference/contribute): AgentsKit is built in the open. Here's how to help — from filing an issue to shipping a new adapter.
- [Launch metrics](https://www.agentskit.io/docs/reference/contribute/launch-metrics): How we measure understanding, activation, use, contribution, and retention for ecosystem launch.
- [Local setup](https://www.agentskit.io/docs/reference/contribute/local-setup): Clone, install, and run the AgentsKit monorepo locally in under 2 minutes.
- [Maintainer expectations](https://www.agentskit.io/docs/reference/contribute/maintainer-expectations): Explicit support ownership and response expectations for ecosystem contributors.
- [Newcomer journey](https://www.agentskit.io/docs/reference/contribute/newcomer-journey): Understand the AgentsKit ecosystem, run a verified demo, and choose a contribution path without private guidance.
- [Package documentation checklist](https://www.agentskit.io/docs/reference/contribute/package-docs): Use this when adding or changing a package or its public API.
- [Recipe and showcase submission](https://www.agentskit.io/docs/reference/contribute/recipe-submission): Submit a verified community showcase or recipe for review without private channels.
- [RFC process](https://www.agentskit.io/docs/reference/contribute/rfc-process): For larger proposals — new packages, breaking changes, or architectural shifts. Keep it lightweight.
- [Roadmap](https://www.agentskit.io/docs/reference/contribute/roadmap): What we're building next, what's open, what's landed.
- [Agent Actions](https://www.agentskit.io/docs/reference/examples/agent-actions): AI agents that generate live, interactive UI — task trackers, dashboards, forms. The agent doesn't just respond with text, it builds working interfaces.
- [Basic Chat](https://www.agentskit.io/docs/reference/examples/basic-chat): The simplest use case — streaming AI conversation with auto-scroll, stop button, and keyboard handling. All in 10 lines with AgentsKit.
- [Code Assistant](https://www.agentskit.io/docs/reference/examples/code-assistant): Streaming code with syntax highlighting. AgentsKit's CodeBlock component renders code beautifully as it streams in.
- [Discord Bot](https://www.agentskit.io/docs/reference/examples/discord-bot): Reference Discord bot wrapping createChatTrigger from @agentskit/runtime. Verifies inbound interactions with Ed25519 public key. No discord.js dependency.
- [DSPy](https://www.agentskit.io/docs/reference/examples/dspy): Baseline prompt vs DSPy-optimized prompt, scored side-by-side against the AgentsKit eval pipeline.
- [Edge (Cloudflare Workers)](https://www.agentskit.io/docs/reference/examples/edge): AgentsKit running on Cloudflare Workers. Demonstrates the sub-50 KB hot path — adapter + ReadableStream, no runtime, no memory, no tools.
- [Embedded (CLI / Raycast / VS Code)](https://www.agentskit.io/docs/reference/examples/embedded): Same agent runs from anywhere. Three host integrations of the identical runtime — Node CLI, Raycast script command, VS Code task.
- [Evaluation Runner](https://www.agentskit.io/docs/reference/examples/eval-runner): Benchmark your agents against test suites. Measure accuracy, latency, and cost with @agentskit/eval.
- [Flow (compileFlow)](https://www.agentskit.io/docs/reference/examples/flow): Live demo of compileFlow — YAML FlowDefinition compiled into a durable DAG, executed with a JSONL step log.
- [Examples](https://www.agentskit.io/docs/reference/examples): Interactive low-level binding demos. For copy-paste code, see Recipes.
- [Ink Example](https://www.agentskit.io/docs/reference/examples/ink): Terminal chat app with tool calling and file-based memory — built with @agentskit/ink.
- [Markdown Chat](https://www.agentskit.io/docs/reference/examples/markdown-chat): Rich formatted responses — headings, tables, code blocks, lists, and blockquotes. AgentsKit's Markdown component renders everything beautifully as it streams.
- [MUI Chat](https://www.agentskit.io/docs/reference/examples/mui-chat): AgentsKit's useChat hook styled with Material UI components. This demo recreates the MUI look — in a real app, you'd use actual MUI imports.
- [Multi-Agent Planning](https://www.agentskit.io/docs/reference/examples/multi-agent): A planner agent that breaks tasks into steps and delegates to specialists. Built with @agentskit/react and the planner skill.
- [Multi-Model Comparison](https://www.agentskit.io/docs/reference/examples/multi-model): Compare responses from different AI models side-by-side. Same input, different adapters — AgentsKit makes it trivial.
- [RAG Chat](https://www.agentskit.io/docs/reference/examples/rag-chat): Retrieval-Augmented Generation — chat with your documents. Show citations, source references, and document context alongside AI responses.
- [RAG Pipeline](https://www.agentskit.io/docs/reference/examples/rag-pipeline): Ingest documents, embed them, and retrieve relevant context during chat. Uses @agentskit/rag with any embedder and vector store.
- [React Example](https://www.agentskit.io/docs/reference/examples/react): Browser chat app with tool calling and localStorage memory — built with @agentskit/react.
- [Runtime Agent](https://www.agentskit.io/docs/reference/examples/runtime-agent): Run a standalone agent from code — no UI required. The runtime executes a ReAct loop: call tools, observe results, decide next action.
- [shadcn Chat](https://www.agentskit.io/docs/reference/examples/shadcn-chat): AgentsKit's useChat hook styled with shadcn/ui patterns. This demo recreates the shadcn look — in a real app, you'd use actual shadcn components.
- [Slack Bot](https://www.agentskit.io/docs/reference/examples/slack-bot): Reference Slack bot wrapping createChatTrigger from @agentskit/runtime. Uses Events API + chat.postMessage REST. No Bolt dependency.
- [Customer Support Bot](https://www.agentskit.io/docs/reference/examples/support-bot): Quick replies, typing indicators, escalation flows. Build production support experiences with AgentsKit's headless components.
- [Teams Bot](https://www.agentskit.io/docs/reference/examples/teams-bot): Reference Microsoft Teams bot wrapping createChatTrigger from @agentskit/runtime. Consumes Bot Framework activities at /api/messages with injectable JWT verifier.
- [Tool Use](https://www.agentskit.io/docs/reference/examples/tool-use): AI assistants that call functions — weather, search, DB queries. Tool calls render as expandable cards.
- [WebLLM (browser-only)](https://www.agentskit.io/docs/reference/examples/webllm): Browser-only chat — the LLM runs 100% in the browser via WebGPU + @mlc-ai/web-llm. No API key, no server-side inference, no telemetry.
- [@agentskit/adapters](https://www.agentskit.io/docs/reference/packages/adapters): 20+ LLM chat + embedder adapters, plus router / ensemble / fallback. Swap providers with one import line.
- [@agentskit/angular](https://www.agentskit.io/docs/reference/packages/angular): Angular service exposing chat state as a Signal + RxJS Observable. Same contract as @agentskit/react.
- [@agentskit/cli](https://www.agentskit.io/docs/reference/packages/cli): The agentskit CLI — init, chat, run, dev, doctor, ai, tunnel, rag, config.
- [@agentskit/core](https://www.agentskit.io/docs/reference/packages/core): Shared contract layer — TypeScript types, headless chat controller, stream helpers. Zero-dep, under 10 KB gzipped.
- [@agentskit/eval/braintrust](https://www.agentskit.io/docs/reference/packages/eval-braintrust): Braintrust scoring pipeline for @agentskit/eval — quality + robustness scorers, CI regression alerts, dataset sync.
- [@agentskit/eval](https://www.agentskit.io/docs/reference/packages/eval): Eval suites + deterministic replay + snapshot testing + prompt diff + CI reporters.
- [@agentskit/ink](https://www.agentskit.io/docs/reference/packages/ink): Terminal chat UI on Ink. Same useChat contract as the React binding.
- [@agentskit/memory](https://www.agentskit.io/docs/reference/packages/memory): Chat memory + vector stores + hierarchical / encrypted / graph / personalization wrappers.
- [@agentskit/observability/langfuse](https://www.agentskit.io/docs/reference/packages/observability-langfuse): Langfuse tracing backend for @agentskit/observability — spans for plan, tool, model, and HITL gates with token/cost/latency capture.
- [@agentskit/observability](https://www.agentskit.io/docs/reference/packages/observability): Trace viewer, signed audit log, cost guard, devtools, token counters.
- [Packages overview](https://www.agentskit.io/docs/reference/packages/overview): Every AgentsKit package at a glance — what it does, when to reach for it, where to read the deep dive.
- [@agentskit/rag](https://www.agentskit.io/docs/reference/packages/rag): Plug-and-play RAG. Chunk, embed, retrieve, rerank, hybrid — plus six document loaders.
- [@agentskit/react-native](https://www.agentskit.io/docs/reference/packages/react-native): React Native / Expo hook. Metro + Hermes safe. Same contract as @agentskit/react.
- [@agentskit/react](https://www.agentskit.io/docs/reference/packages/react): React hooks + headless chat components driving the shared ChatController.
- [Roadmap](https://www.agentskit.io/docs/reference/packages/roadmap): Per-package stability status, current version, and what each package needs to reach v1.0.
- [@agentskit/runtime](https://www.agentskit.io/docs/reference/packages/runtime): Standalone agent runtime. ReAct loop, durable execution, multi-agent topologies, speculative execution, background agents.
- [@agentskit/sandbox](https://www.agentskit.io/docs/reference/packages/sandbox): Secure code execution (E2B / WebContainer) + mandatory-sandbox policy wrapper.
- [@agentskit/skills](https://www.agentskit.io/docs/reference/packages/skills): Ready-made personas + composition + marketplace registry with semver.
- [@agentskit/solid](https://www.agentskit.io/docs/reference/packages/solid): Solid hook. Same contract as @agentskit/react.
- [@agentskit/statechart](https://www.agentskit.io/docs/reference/packages/statechart): Deterministic, serializable interaction state without a UI or execution-engine dependency.
- [@agentskit/svelte](https://www.agentskit.io/docs/reference/packages/svelte): Svelte 5 chat store. Same contract as @agentskit/react.
- [@agentskit/templates](https://www.agentskit.io/docs/reference/packages/templates): Authoring toolkit for AgentsKit extensions — validated Tool, Skill, Adapter factories + secure on-disk scaffolds.
- [@agentskit/tools](https://www.agentskit.io/docs/reference/packages/tools): Built-in tools + 20 third-party integrations + bidirectional MCP bridge.
- [@agentskit/vue](https://www.agentskit.io/docs/reference/packages/vue): Vue 3 composable + ChatContainer component. Same contract as @agentskit/react.
- [Adapter contract tests](https://www.agentskit.io/docs/reference/recipes/adapter-contract-tests): Verify any adapter against the ADR 0001 invariants A1–A10 with the shared test harness.
- [Adapter ensemble](https://www.agentskit.io/docs/reference/recipes/adapter-ensemble): Send one request to N models in parallel, aggregate the answers into a single output.
- [Adapter router](https://www.agentskit.io/docs/reference/recipes/adapter-router): Auto-pick an adapter per request based on cost, latency, capabilities, or a custom classifier.
- [agentskit ai — natural-language agent generator](https://www.agentskit.io/docs/reference/recipes/agentskit-ai): Describe an agent in plain English, get a typed AgentSchema + tool stubs + runtime wiring.
- [Signed audit log](https://www.agentskit.io/docs/reference/recipes/audit-log): Hash-chained, HMAC-signed audit log for SOC 2 / HIPAA friendly evidence — tamper-evident and authenticated.
- [Auto-summarization](https://www.agentskit.io/docs/reference/recipes/auto-summarize): Wrap any ChatMemory so it compacts oldest messages into a summary whenever stored tokens exceed a budget.
- [Background agents (cron + webhooks)](https://www.agentskit.io/docs/reference/recipes/background-agents): Run agents on a schedule or in response to incoming webhooks, without pulling in a job-queue dependency.
- [Recipe: Bail / Qwen routing](https://www.agentskit.io/docs/reference/recipes/bail-qwen-routing): Use the bail (Alibaba DashScope) / qwen adapter alongside Western providers, with cost-aware routing for Asia-Pacific traffic.
- [Browser-only chat (WebLLM / WebGPU)](https://www.agentskit.io/docs/reference/recipes/browser-only-webllm): Ship a chat agent that runs 100% in the user's browser — no server, no API key, no telemetry. Privacy contract holds because no inference data ever leaves the device.
- [Code reviewer agent](https://www.agentskit.io/docs/reference/recipes/code-reviewer): An agent that reads a git diff, runs the test suite, and posts a structured review.
- [Confirmation-gated tool](https://www.agentskit.io/docs/reference/recipes/confirmation-gated-tool): A dangerous tool the runtime refuses to execute without explicit human approval.
- [Cost guard](https://www.agentskit.io/docs/reference/recipes/cost-guard): Enforce a dollar budget per run. Tokens → cost → abort, all via contract primitives.
- [Cost-guarded chat](https://www.agentskit.io/docs/reference/recipes/cost-guarded-chat): A chat that aborts the run when token usage exceeds your budget — using only an observer.
- [Custom adapter](https://www.agentskit.io/docs/reference/recipes/custom-adapter): Wrap any LLM API as an AgentsKit adapter. Plug-and-play with the rest of the kit in 30 lines.
- [Deterministic replay](https://www.agentskit.io/docs/reference/recipes/deterministic-replay): Record a real agent session once, then replay it bit-for-bit in tests and bug repros — no more flaky LLM traces.
- [Devtools server](https://www.agentskit.io/docs/reference/recipes/devtools-server): Expose a live feed of agent events so any browser extension or custom dashboard can inspect a running agent.
- [Discord bot](https://www.agentskit.io/docs/reference/recipes/discord-bot): A Discord bot powered by AgentsKit. Replies in threads, calls tools, remembers per-channel.
- [Document loaders](https://www.agentskit.io/docs/reference/recipes/doc-loaders): Fetchers for URL, GitHub, Notion, Confluence, Drive, PDF, and cloud stores into your RAG pipeline.
- [Durable execution (Temporal-style)](https://www.agentskit.io/docs/reference/recipes/durable-execution): Wrap side-effectful steps so crashes, deploys, and retries replay from a step log instead of starting over.
- [Edit + regenerate messages](https://www.agentskit.io/docs/reference/recipes/edit-and-regenerate): Let users correct a prompt, edit the model's answer, or re-run any assistant turn — with correct truncation and streaming.
- [Encrypted memory](https://www.agentskit.io/docs/reference/recipes/encrypted-memory): Client-side AES-GCM encryption for any ChatMemory — keys never leave the caller, backing store only sees ciphertext.
- [Eval suite for an agent](https://www.agentskit.io/docs/reference/recipes/eval-suite): Score an agent's quality in CI. A test suite for agents, not for code.
- [Evals in CI](https://www.agentskit.io/docs/reference/recipes/evals-ci): Run agent evals on every PR, fail builds below a minimum accuracy, surface results in the PR UI.
- [Fallback chain](https://www.agentskit.io/docs/reference/recipes/fallback-chain): Try adapters in order — on error, fall through to the next without duplicating tool calls.
- [Recipe: Figma design extraction](https://www.agentskit.io/docs/reference/recipes/figma-design-extraction): Pull frames from a Figma file, render them as PNGs, and feed them to a vision-capable model for design QA.
- [Vue / Svelte / Solid / React Native / Angular](https://www.agentskit.io/docs/reference/recipes/framework-adapters): One package per framework. Same ChatReturn contract as @agentskit/react — pick the binding that matches your stack.
- [Generative UI + artifacts](https://www.agentskit.io/docs/reference/recipes/generative-ui): A typed JSON schema the agent emits, a framework-agnostic element tree, plus rich artifacts (code, markdown, HTML, charts).
- [Graph memory](https://www.agentskit.io/docs/reference/recipes/graph-memory): Non-linear memory for entities and relationships. Backs anything from in-memory Maps to Neo4j.
- [Hierarchical memory (MemGPT-style)](https://www.agentskit.io/docs/reference/recipes/hierarchical-memory): Three tiers — working, recall, archival — so long-running agents don't lose context and don't blow the window.
- [Human-in-the-loop approvals](https://www.agentskit.io/docs/reference/recipes/hitl-approvals): Pause an agent at a named gate, persist the decision, resume deterministically.
- [Recipe: HubSpot / Airtable / Shopify daily pulse](https://www.agentskit.io/docs/reference/recipes/hubspot-airtable-shopify-pulse): A single agent that pulls today's signals from your CRM, ops board, and storefront — then drafts a Slack digest.
- [Recipes](https://www.agentskit.io/docs/reference/recipes): Copy-paste solutions grouped by theme. Every recipe end-to-end, runs as written.
- [Provider integrations](https://www.agentskit.io/docs/reference/recipes/integrations): Ready-made tools for GitHub, Linear, Slack, Notion, Discord, Gmail, Google Calendar, Stripe, Postgres, and S3.
- [Recipe: Jira triage agent](https://www.agentskit.io/docs/reference/recipes/jira-triage): An agent that watches a Jira project, classifies new tickets, and assigns them — using the jira integration.
- [Mandatory tool sandbox](https://www.agentskit.io/docs/reference/recipes/mandatory-sandbox): Enforce a sandbox policy across every tool the agent can call — allow-list, deny-list, require-sandbox, per-tool validators.
- [MCP bridge (bidirectional)](https://www.agentskit.io/docs/reference/recipes/mcp-bridge): Consume any MCP server as AgentsKit tools, or expose your AgentsKit tools to MCP hosts — over stdio or any transport.
- [Scraping, voice, maps, browser agent](https://www.agentskit.io/docs/reference/recipes/more-integrations): Firecrawl + Jina Reader, OpenAI Images + ElevenLabs + Whisper + Deepgram, Nominatim + OpenWeatherMap + CoinGecko, and a BYO-Playwright browser agent.
- [More provider adapters](https://www.agentskit.io/docs/reference/recipes/more-providers): Mistral, Cohere, Together, Groq, Fireworks, OpenRouter, Hugging Face, LM Studio, vLLM, llama.cpp — one line each.
- [Multi-agent topologies](https://www.agentskit.io/docs/reference/recipes/multi-agent-topologies): Ready-made supervisor, swarm, hierarchical, and blackboard builders — four ways to combine agents into one.
- [Unified multi-modal](https://www.agentskit.io/docs/reference/recipes/multi-modal): One API for text, image, audio, video, and file inputs — regardless of provider.
- [Open specs — A2A, Manifest, Eval Format](https://www.agentskit.io/docs/reference/recipes/open-specs): Three small, versioned specs so agents, skill packs, and eval datasets travel across tools.
- [PDF Q&A](https://www.agentskit.io/docs/reference/recipes/pdf-qa): Ask questions about a local PDF file. Extract, chunk, embed, retrieve, answer.
- [Persistent memory across sessions](https://www.agentskit.io/docs/reference/recipes/persistent-memory): A chat that remembers yesterday's conversation. SQLite-backed. Works in 5 lines.
- [Personalization](https://www.agentskit.io/docs/reference/recipes/personalization): Persisted user profile that conditions every agent response.
- [PII redaction](https://www.agentskit.io/docs/reference/recipes/pii-redaction): Strip emails, phone numbers, SSNs, and other PII from messages before they hit the model or your logs.
- [Progressive tool calls](https://www.agentskit.io/docs/reference/recipes/progressive-tool-calls): Start executing a tool before the model finishes streaming its arguments.
- [Prompt diff](https://www.agentskit.io/docs/reference/recipes/prompt-diff): Git blame for prompts — find which prompt change is responsible for an output change.
- [A/B prompts with feature flags](https://www.agentskit.io/docs/reference/recipes/prompt-experiments): Ship multiple prompts, route users deterministically, measure which wins.
- [Prompt injection detector](https://www.agentskit.io/docs/reference/recipes/prompt-injection): Score incoming text for injection attempts — heuristics + optional model classifier (Llama Guard, Rebuff).
- [Prompt snapshot testing](https://www.agentskit.io/docs/reference/recipes/prompt-snapshots): Jest-style snapshot tests for prompts, with semantic tolerance so small wording drift doesn't break CI.
- [Chat with RAG](https://www.agentskit.io/docs/reference/recipes/rag-chat): A streaming React chat that answers from your own documents. Vector store, embeddings, retrieval, hooked up in 30 lines.
- [RAG reranking + hybrid search](https://www.agentskit.io/docs/reference/recipes/rag-reranking): Wrap your vector retriever with BM25 hybrid scoring and pluggable rerankers (Cohere Rerank, BGE, etc.).
- [Rate limiting](https://www.agentskit.io/docs/reference/recipes/rate-limiting): Token-bucket rate limits keyed by user / IP / API key with per-tier bucket config.
- [Replay a session against a different model](https://www.agentskit.io/docs/reference/recipes/replay-different-model): Re-run a recorded cassette through any adapter to compare quality, latency, or cost without touching production traffic.
- [Multi-agent research team](https://www.agentskit.io/docs/reference/recipes/research-team): A planner that delegates to a researcher and a writer. Real multi-agent in 30 lines.
- [Schema-first agents](https://www.agentskit.io/docs/reference/recipes/schema-first-agent): Define your agent in YAML or JSON and get a typed AgentSchema you can feed the runtime.
- [Self-debug tool](https://www.agentskit.io/docs/reference/recipes/self-debug): On tool error, let the agent read the error + schema and draft corrected arguments for a retry.
- [Recipe: Sentry incident bot](https://www.agentskit.io/docs/reference/recipes/sentry-incident-bot): Watch a Sentry project for new errors; cluster, summarise, and assign owners.
- [Wrap a non-streaming endpoint](https://www.agentskit.io/docs/reference/recipes/simulate-stream): Turn a one-shot provider into a streaming adapter so UIs see identical ergonomics.
- [Skill marketplace + ready-made skills](https://www.agentskit.io/docs/reference/recipes/skill-marketplace): Publish + install versioned skills through a registry. Four new ready-made skills.
- [Speculative execution](https://www.agentskit.io/docs/reference/recipes/speculative-execution): Run the same request across N adapters in parallel, keep the winner, abort the losers.
- [Recipe: Scaffolding with @agentskit/templates](https://www.agentskit.io/docs/reference/recipes/templates-cookbook): Generate a build-ready AgentsKit package skeleton — tool, skill, adapter, embedder, memory, browser-adapter, or flow — in a few lines of Node.
- [Time-travel debug](https://www.agentskit.io/docs/reference/recipes/time-travel-debug): Step through a recorded agent session, rewrite a tool result, and replay from that point forward.
- [Token budget compiler](https://www.agentskit.io/docs/reference/recipes/token-budget): Declare a token budget, let AgentsKit trim messages and summarize history to fit.
- [Tool composer](https://www.agentskit.io/docs/reference/recipes/tool-composer): Chain N tools into a single macro tool — a fixed recipe the model can invoke with one schema.
- [Local trace viewer](https://www.agentskit.io/docs/reference/recipes/trace-viewer): Persist agent spans to disk and open a self-contained HTML waterfall — Jaeger-style, no server required.
- [Trigger adapters (email / teams / postgres-cdc)](https://www.agentskit.io/docs/reference/recipes/triggers-adapters): Reference adapter snippets that wrap heavy drivers (nodemailer, imapflow, botbuilder, pg-logical-replication) into the driver-free contracts the AgentsKitOS triggers package consumes.
- [Vector memory adapters](https://www.agentskit.io/docs/reference/recipes/vector-adapters): Drop-in VectorMemory for pgvector, Pinecone, Qdrant, Chroma, and Upstash Vector.
- [Recipe: vector filter helpers](https://www.agentskit.io/docs/reference/recipes/vector-filter-helpers): Use matchesFilter to evaluate vector-store filter predicates outside an adapter, and pair with postgresWithRoles for row-level security.
- [Virtualized memory](https://www.agentskit.io/docs/reference/recipes/virtualized-memory): Transparently handle giant conversations by keeping a hot window active and paging the rest.
- [A2A — Agent-to-Agent](https://www.agentskit.io/docs/reference/specs/a2a): JSON-RPC 2.0 contract for one agent to invoke another across process or network.
- [AgentSchema](https://www.agentskit.io/docs/reference/specs/agent-schema): Typed, validated definition of an agent — adapter, tools, skills, memory, rag, observers.
- [Eval format](https://www.agentskit.io/docs/reference/specs/eval-format): Portable JSON for eval datasets + run results. Tool-agnostic.
- [Generative UI](https://www.agentskit.io/docs/reference/specs/generative-ui): Schema-driven renderers. LLM outputs typed JSON; UI reflects it.
- [Open specs](https://www.agentskit.io/docs/reference/specs): Three portable JSON contracts — Agent-to-Agent, Skill+Tool Manifest, Eval Format.
- [Manifest](https://www.agentskit.io/docs/reference/specs/manifest): Portable packaging for skills + tools. MCP-compatible tool entries.

## Optional

- [Full markdown index](https://www.agentskit.io/llms-full.txt): complete text of every docs page in one file
- [Sitemap](https://www.agentskit.io/sitemap.xml): machine-readable URL list
- [GitHub](https://github.com/AgentsKit-io/agentskit): source + issues