Open Source · Apache 2.0

Your AI needs
a team,
not a prompt.

Deploy persistent AI teams where agents communicate, share knowledge, and coordinate autonomously. TypeScript. Self-hosted. Any LLM.

terminal
~
~
Compatible with
Anthropic OpenAI Ollama Claude Code OpenClaw + OpenRouter, Together, Groq, Fireworks, DeepSeek…
2,124+Tests
135API Routes
8Templates
5Adapters
26DB Tables
18Job Types
01 The problem

Every framework claims "multi-agent."
None of them ship real communication.

Sequential Pipelines

Pass strings between agents in a fixed order. No real messaging, no knowledge sharing, no ability for agents to coordinate autonomously.

Group Chat

Put agents in a shared room where everyone sees everything. No scoped access, no structured channels, no privacy between teams.

State Machines

Route through rigid graphs. Great for deterministic workflows, but agents can't decide who to talk to or when to escalate.

Ticket Systems

File tickets between agents like a project management tool. No real-time messaging, no broadcast channels, no shared knowledge base.

02 How it works

Three steps to a running AI team.

01

Deploy a team

Pick a template. Run one command. Three specialized agents spin up with their own identity, memory, and scheduled heartbeats.

02

Agents communicate

Direct messaging, broadcast channels, escalation chains. A security agent publishes findings — every subscribed agent gets it automatically.

03

Knowledge flows

Research discovers a trend. Writes to shared knowledge. Content creates around it. Social amplifies it. CEO coordinates. All autonomous.

03 Features

What no other framework has.

Structured Communication

11 action types. Broadcast channels with role-based permissions. Escalation chains. Cross-branch routing.

11 ACTION TYPES · ROLE-BASED ACL · PUB/SUB TOPICS

Shared Knowledge

Three-level knowledge store — private notes, team wiki, company wiki. Auto-expiry, tag search, scoped access control.

3 SCOPE LEVELS · AUTO-EXPIRY · TAG SEARCH

Intelligent Context

Every invocation builds a structured briefing from 22 data sources, priority-truncated to fit the token budget.

22 DATA SOURCES · PRIORITY TRUNCATION · TOKEN-AWARE

Self-Healing Coordination

Circular task dependencies are detected and resolved automatically. Agents never get stuck. No human intervention needed.

DEADLOCK DETECTION · AUTO-RESOLVE · ZERO INTERVENTION
04 Templates

Pre-built AI teams. Deploy in one command.

Security & Code Review

OWASP top 10, dependency CVE monitoring, automated pentesting, posture reports.

security-code-review

DevOps & Incident Response

Health checks, automated incident response, deployment management, post-mortems.

devops-incident-response

Research & Intelligence

Multi-source monitoring, trend synthesis, competitive analysis, intelligence briefings.

research-intelligence

Content Production

Content calendar, AI-assisted drafting, multi-platform distribution, performance tracking.

content-production

OSS Management

Issue triage, PR review, community engagement, release coordination, onboarding.

oss-management

Deal Flow & Sales

Lead discovery, company enrichment, personalized outreach, pipeline management.

deal-flow

Social Media Growth

Campaign management, content scheduling, engagement analytics, audience growth.

social-media-growth

Head Agent (CEO)

Coordinates all branches, routes cross-branch tasks, monitors health, onboards teams.

head-agent
05 Adapters

Works with your AI.

Five adapters shipped. Mix providers per agent. CEO on Claude, workers on Ollama.

Registered Adapters 5 active
HTTP
OpenAI-compatible
Anthropic
Claude models
Ollama
Local models
Claude Code
CLI adapter
OpenClaw
Gateway
06 Quickstart

Running in 60 seconds.

terminal
# Initialize a new flock
$ npx flockrun init

# Start the server
$ npm start

# Open dashboard at localhost:3000
# Pick a template. Deploy. Watch agents coordinate.

Zero external dependencies in dev mode. SQLite built-in. No Redis, no Docker, no Kubernetes. Just Node.js.

07 Compare

How FlockRun compares.

Capability
FlockRun
Others
Real-time agent messaging
11 action types
String passing or shared chat
Scoped knowledge base
3-level store
Basic or none
Multi-source context building
22 sources
Basic context passing
Self-healing coordination
Auto-resolve
Retries only
Pre-built team templates
8 templates
Few or none
Adapter-agnostic
5 adapters
Varies by framework
Self-hosted (no cloud lock-in)
SQLite + Node
Some self-hosted options
TypeScript-native
Mostly Python

Built in the open.

Months of iterative development. Production-deployed. Not a weekend project. Every line of code available to read, audit, and contribute to.

2,124+integration tests
28,000+lines of TypeScript
Apache 2.0open source license