Recursiv

Build, run and govern autonomous agents

What is Recursiv?

Recursiv is the platform for building, running and governing autonomous agents. One SDK and one API key give you the whole backend: projects, Postgres, storage, deploys, auth, billing, agents, memory, orchestration, hundreds of integrations, plus the identity, guardrails, observability and verification to govern every action. It is model agnostic, so you bring any model and swap it anytime.

Use it from the TypeScript SDK, REST API, CLI, or MCP server. Humans get a direct developer platform. Agents get a deterministic tool surface with clear limits, scopes and machine-readable docs.

App, CLI, or agent
-> @recursiv/sdk, REST, MCP, WebSocket
-> projects, Postgres, storage, deploy, auth, billing
-> agents, memory, orchestration, integrations
-> identity, guardrails, observability, verification

Use Recursiv when

  • You want a product backend without stitching together auth, database, storage, deploy, social, and agent infrastructure.
  • You want agents to build, inspect, and ship inside scoped infrastructure instead of loose shell scripts.
  • You need social primitives such as posts, feeds, chat, communities, profiles, and follow graph.
  • You need operational primitives such as webhooks, scheduled jobs, billing, observability, and self-hosting.
  • You want the same platform available to app code, humans in a terminal, and MCP-compatible coding agents.

What you get

AreaCurrent primitives
Projects and deployProject CRUD, branch deploys, deployment status, deployment logs, sandbox lifecycle
DatabasesManaged Postgres, idempotent provisioning, credentials API, parameterized SQL queries
StorageBuckets, folders, presigned upload/download URLs, object listing, deletes
AI agentsAgent CRUD, streaming chat, project permissions, inbox, conversations, discovery, scheduled work
Swarms and dispatcherTask queues, claims, stale/stuck detection, swarms, schedules, outcomes, signals
Memory and brainFacts, decisions, search, project tasks, milestones, settings, usage
Social and realtimePosts, tags, communities, profiles, follow graph, DMs, group chat, WebSocket hooks
PlatformAuth, API keys, organizations, billing, notifications, webhooks, jobs, settings
Agent toolingMCP server, scoped tools, setup resources, self-evaluation, llms.txt

Quick start

1. Install the SDK

Node.js 18 or newer is required.

$npm install @recursiv/sdk

2. Set your API key

$export RECURSIV_API_KEY=sk_live_...

The SDK also accepts apiKey in the constructor and falls back to SOCIAL_DEV_API_KEY for older deployments.

3. Make the first call

1import { Recursiv } from '@recursiv/sdk';
2
3const r = new Recursiv(); // reads RECURSIV_API_KEY
4
5const { data: db } = await r.databases.ensure({
6 project_id: 'proj_1',
7 name: 'main',
8});
9
10const { data: result } = await r.databases.query({
11 project_id: 'proj_1',
12 sql: 'SELECT NOW() as time',
13});
14
15console.log(db.id, result.rows);

4. Try the anonymous sandbox

No signup or API key is required.

1import { Recursiv } from '@recursiv/sdk';
2
3const r = new Recursiv({ anonymous: true });
4
5const { data, meta } = await r.sandbox.execute({
6 code: 'console.log(1 + 1)',
7 language: 'typescript',
8});
9
10console.log(data.output); // "2\n"
11console.log(`${meta.remaining_executions} executions remaining today`);

Anonymous sandbox limit: 10 executions per IP per day.

Choose your integration

Use thisWhen you wantStart here
TypeScript SDKTyped app code, scripts, server routes, or agent-generated codeInstallation
CLIScaffold, authenticate, inspect, and generate deploy config from a terminalCLI
MCPGive Claude Code, Claude Desktop, Cursor, Codex, or another MCP client toolsMCP
REST APIUse Recursiv from any language or HTTP clientAPI Reference
Agent docsGive an agent the full docs index and page-level Markdownllms.txt

What do you want to build?

GoalWhat Recursiv gives youGuide
AI-native appDatabase + AI agent + storage + auth + billingBuild an AI-native app
Autonomous agentAgent with its own Postgres, memory and code executionBuild an autonomous agent
Social platformPosts, communities, chat, profiles, follow graphBuild a social platform
Code playgroundSandboxes, deploy, AI coding assistant, live previewBuild a code playground
White-label SaaSBYOK Stripe + branded auth + branded email + your infrastructureDeploy a white-label platform
Any app that needs a backendAuth + database + storage + billing + emailQuickstart

Agent rules of thumb

  • Use r.databases for structured app data. Do not store app records as JSON blobs in posts.
  • Use r.posts, r.communities, r.chat, and r.profiles for social product surfaces.
  • Use r.projects.createSandbox() and r.projects.executeCode() for project-scoped code execution.
  • Use r.sandbox.execute() only for anonymous, temporary code execution.
  • Use r.agents.chatStream() for token streaming in Node.js and browsers.
  • Use r.agents.chatStreamText() or the React Native guide when ReadableStream is unavailable.
  • Keep API keys server-side in browser apps.

Next steps