AI agent backend for SaaS builders
Give every customer their own AI agent without building a runtime. FastClaw handles agents, end-user isolation, sandboxed tools and usage metering over an API.
OpenAI-compatible API
Call /v1/chat/completions with streaming, tool-call progress and per-turn token usage. Existing OpenAI clients work unchanged.
End-users without sign-up
Pass X-Fastclaw-End-User and FastClaw mints an isolated namespace per user — sessions, memory and files never leak between your customers.
Agents as an API resource
Create, update, list and delete agents on /v1/agents, upload SOUL.md and other system files, and filter by your own metadata.
Sandboxed by default when hosted
With FASTCLAW_DEPLOY=hosted every exec and file call is locked inside an E2B or Docker sandbox, so untrusted users never touch the host.
Billing hooks, not a billing system
Export usage events per account, put an account on billing hold, and mint one-time login links — then charge with the payment stack you already have.
Files come back with the reply
Files an agent writes during a turn are returned with download URLs scoped to the caller’s namespace, or inline as data URLs for small files.
Skip the runtime, ship the feature
Adding "an AI agent that can actually do things" to a product usually means building an agent loop, tool execution, a sandbox, session storage, memory and per-user isolation before you write a single line of product code. FastClaw is that layer, packaged as one Go binary you run next to your app.
Your app keeps its own users, auth and billing. It talks to FastClaw over the runtime API: one account per customer, a user-scope API key, and an end-user ID per person. FastClaw does the rest — the agent loop, model routing, sandboxed tools and persistent sessions.
A typical integration
- Create an account and a
userAPI key for your product. - Create one or more agents with
POST /v1/agentsand upload theirSOUL.md. - From your backend, call
/v1/chat/completionswithX-Fastclaw-End-Userset to your user's ID. Stream the reply to your UI. - Pull
GET /api/admin/usage/eventson a cursor and bill however you like.
Every FastClaw instance serves an integration guide at /skills/agent-integration/SKILL.md, written so a coding agent can wire your app up for you.
Grows from SQLite to a cluster
Start on a laptop with SQLite. When traffic grows, point storage at Postgres, add an S3-compatible object store and Redis, and run several replicas behind a load balancer. Isolate tool execution in E2B or Docker. Compare this approach with an app builder in FastClaw vs Dify, or check the licensing options on pricing.
Works with
All Integrations
OpenAI
Run FastClaw agents on OpenAI models — set a deployment-wide default, override per agent, and expose your agents through an OpenAI-compatible API.
E2B
Run every FastClaw agent command in an isolated E2B cloud sandbox — skills and workspace hydrated in, files synced back after each call.
PostgreSQL
Swap FastClaw's default SQLite file for PostgreSQL to run multiple pods behind a load balancer, with automatic schema migrations.
Frequently asked questions
More Use cases
All Use casesDevelopers
Run your own AI agents on your laptop or server with one Go binary. Any model, local or hosted, reachable from the terminal, the web or your chat apps.
Teams
Deploy shared AI agents inside the chat tools your team already uses. Self-hosted, with per-person sessions, shared skills and scheduled reports.
Communities
Run an AI agent for your community on Telegram, Discord or LINE. Every member gets their own session, and untrusted chatters stay inside a sandbox.
Try it on your own agents
Start free in FastClaw Cloud, or self-host the single binary in a minute.