DISCOVER THE FUTURE OF AI AGENTSarrow_forward

AgentFactory

calendar_todayAdded Apr 23, 2026
categoryAgent & Tooling
codeOpen Source
TypeScriptNode.jsWorkflow AutomationMulti-Agent SystemModel Context ProtocolAI AgentsAgent FrameworkAgent & ToolingDeveloper Tools & CodingAutomation, Workflow & RPAProtocol, API & Integration

The open-source software factory — multi-agent fleet management for coding agents, automating the full pipeline from backlog to merge.

AgentFactory is an open-source software factory built by Rensei AI, focused on fleet-level orchestration of multiple coding agents. Built with TypeScript using a pnpm monorepo + Turborepo architecture, it schedules Claude, Codex, Amp, Spring AI, and A2A-compatible agents through a unified AgentProvider interface, with each agent working in an isolated Git worktree in parallel.

The core orchestration pipeline automatically advances issues through five stages: Backlog → Development → QA → Acceptance → Merge. A Workflow Governor enables upstream automation from Research to backlog creation to development. The system features intelligent routing based on Thompson Sampling multi-armed bandit algorithm, automatically learning optimal provider allocation across work types, with 9-level resolution cascade for dynamic provider selection.

For quality assurance, AgentFactory integrates a code intelligence engine powered by Tree-sitter AST + BM25, providing code search, symbol lookup, repository maps, cross-package dependency validation, and duplicate code detection. Quality gates use baseline capture and metric ratcheting to prevent regressions. The merge queue supports local, GitHub-native, Mergify, and Trunk providers.

On the protocol layer, the project implements A2A Protocol v0.3.0 (as both client and server) and exposes fleet capabilities to MCP-aware clients like Claude Desktop via an MCP Server. Runtime reliability is ensured through heartbeat monitoring, .agent/ directory state persistence, automatic crash recovery, and configurable timeouts. Distributed mode scales via Redis work queues across webhook servers and multiple worker nodes.

The project deeply integrates with Linear as the primary issue data source and state hub, supporting webhook ingestion, automatic state transitions, and sub-issue coordination. It provides CLI tools (af-orchestrator, af-linear), a programmatic API, one-click scaffolding, and quick deployment to Vercel/Railway. Released under the MIT license, the latest version is v0.8.51.

Core Monorepo Packages

  • @renseiai/agentfactory: Core orchestrator, Provider abstraction, crash recovery
  • @renseiai/agentfactory-server: Redis work queue + session storage + Worker pool
  • @renseiai/agentfactory-nextjs: Next.js route handlers + Webhook middleware
  • @renseiai/agentfactory-dashboard: Fleet management Dashboard UI
  • @renseiai/agentfactory-code-intelligence: Tree-sitter AST + BM25 search + PageRank repo maps
  • @renseiai/agentfactory-mcp-server: MCP Server (Streamable HTTP + STDIO)

Quick Start

npx @renseiai/create-agentfactory-app my-agent
cd my-agent
cp .env.example .env.local
pnpm install && pnpm dev
pnpm worker

Unconfirmed: Whether Rensei AI has a standalone website; support for non-Linear scenarios (e.g., GitHub Issues) is not documented; specific version or model requirements for each Provider are not specified.

Related Projects

View All arrow_forward

STAY UPDATED

Get the latest AI tools and trends delivered straight to your inbox. No spam, just intelligence.

rocket_launch