Production-proven orchestration layer

Multi-Agent Development Pipelines That Actually Ship Code

33 stories shipped across 3 production repos. 6 coordinated agents. Zero human intervention. Drop into any AI-powered IDE or coding assistant.

Integrate Antfarm See the Evidence
33 Stories Shipped
3 Production Repos
0 Regressions
6 Coordinated Agents
MIT Licensed

The orchestration layer every AI company needs

Every AI company is building multi-agent dev tools right now. Antfarm already works.

For AI Companies

Drop-in agent pipeline for your platform

Building multi-agent coding tools? Antfarm's battle-tested pipeline handles planning, implementation, testing, and code review autonomously. Integrate it into your IDE, copilot, or agent platform.

For Platform Teams

Agents that review each other's work

Your agents write code but who reviews it? Antfarm's 6-agent pipeline includes dedicated verification, testing, and review stages -- catching bugs before they reach production.

For Open Source

Fork it. Define your workflows. Ship.

Fork the agent-template, define your workflows, run against any codebase. MIT licensed. Works with any LLM backend. No vendor lock-in.

Claude Code multi-agent Codex agent pipelines Copilot Workspace Agent-powered IDEs v0 / AI dev platforms

Three pipelines. Zero hand-holding.

Each workflow is a deterministic sequence of agent steps. Same workflow, same order, every time. Not "hopefully the agent remembers to test."

+
feature-dev 7 agents

Drop in a feature request. Get back a tested PR. The planner decomposes your task into stories. Each story gets implemented, verified, and tested in isolation. Failures retry automatically.

plan setup implement verify test PR review
security-audit 7 agents

Point it at a repo. Get back a security fix PR with regression tests. Scans for vulnerabilities, ranks by severity, patches each one, then re-audits after all fixes are applied.

scan prioritize setup fix verify test PR
bug-fix 6 agents

Paste a bug report. Get back a fix with a regression test. Triager reproduces it, investigator finds root cause, fixer patches, verifier confirms. Zero babysitting.

triage investigate setup fix verify PR

How agents work together

YAML + SQLite + cron. No Redis, no Kafka, no container orchestrator. Each agent runs in a fresh session with clean context.

Task Description | v +-------------------+ | Planner | Decomposes task into ordered user stories (max 20) +-------------------+ | v +-------------------+ | Setup | Creates branch, discovers build/test commands, baseline +-------------------+ | v +-----------------------------------------------------+ | Story Execution Loop | | | | +-------------------+ +-------------------+ | | | Developer | --> | Verifier | | | +-------------------+ +-------------------+ | | ^ | | | | STATUS: retry | | | +--------------------------+ | | | | For each story: | | Developer implements + writes tests | | Verifier checks acceptance criteria | | Pass -> next story Fail -> retry (max 2) | +-----------------------------------------------------+ | v +-------------------+ | Tester | Integration/E2E testing across all stories +-------------------+ | v +-------------------+ | PR Creator | Creates pull request with summary +-------------------+ | v +-------------------+ | Reviewer | Reviews PR, approves or requests changes +-------------------+ | v Done -- Tested, reviewed PR ready to merge

Why it actually works

Deterministic Workflows

Same workflow, same steps, same order. Every run is reproducible. No hoping the agent remembers to test.

Agents Verify Each Other

The developer does not mark their own homework. A separate verifier checks every story against acceptance criteria.

Fresh Context, Every Step

Each agent gets a clean session. No context window bloat. No hallucinated state from 50 messages ago.

Retry and Escalate

Failed steps retry automatically. If retries exhaust, it escalates to you. Nothing fails silently.

Minimal by Design

YAML + SQLite + cron. TypeScript CLI with zero external dependencies. Runs wherever OpenClaw runs.

Built on Ralph Loop

Memory persists through git history and progress files. The autonomous loop pattern from Ralph, scaled to multi-agent workflows.

Six agents. Six specializations.

Each agent has its own workspace with identity files: AGENTS.md, SOUL.md, and IDENTITY.md. No ambiguity about who does what.

P
Planner analysis

Decomposes tasks into ordered user stories with verifiable acceptance criteria. Maximum 20 stories per run.

S
Setup coding

Creates branch, discovers build/test commands, establishes baseline. Prepares the workspace for development.

D
Developer coding

Implements stories one at a time, writes tests, commits with structured messages. Works in tight loop with Verifier.

V
Verifier verification

Checks each story against acceptance criteria, runs tests, security checks. Approves or sends feedback for retry.

T
Tester testing

Integration and E2E testing after all stories are implemented. Catches cross-story regressions and edge cases.

R
Reviewer analysis

Reviews the PR, approves or requests changes with actionable feedback. Final quality gate before merge.

Fits where your agents already live

Antfarm is not a replacement for your agent platform. It is the orchestration layer that makes your agents work as a team.

Claude Code

Antfarm workflows as structured multi-agent sessions. Each agent gets a clean Claude Code context with AGENTS.md instructions.

claude → antfarm pipeline

GitHub Copilot

Antfarm's review pipeline as a Copilot Workspace extension. Agents verify, test, and review Copilot-generated code before it merges.

copilot → antfarm review

MCP-Compatible Agents

Any agent that speaks MCP can join an Antfarm workflow. Define tools, context, and handoff points in YAML.

mcp agent → antfarm orchestration

Standalone CLI

No platform required. Point Antfarm at any repo and run deterministic agent workflows from your terminal.

antfarm run feature-dev "implement user auth"

One command. Zero infrastructure.

Requires Node.js 22+ and OpenClaw. No Docker, no queues, no external services.

Install from GitHub

curl -fsSL https://raw.githubusercontent.com/snarktank/antfarm/v0.5.1/scripts/install.sh | bash

or tell your OpenClaw agent:

install github.com/snarktank/antfarm

then run:

antfarm workflow list

33 stories. 3 repos. 0 regressions.

Not a demo. Not a benchmark. Real production Solana repositories, autonomously planned, implemented, tested, and reviewed. February 2026.

tribeca-dao
9 stories

Anchor 0.30 migration, IDL regeneration, TS SDK migration, demo lifecycle. Every commit agent-authored.

View agent commits →

grape-art
11 stories

Full dependency modernization, Parcel build fix, marketplace demo. From broken to building autonomously.

View agent commits →

port-lending
13 stories

Rust dependency updates, cargo build-sbf, test restoration, TS SDK. The largest autonomous run -- 13 stories, zero failures.

View agent commits →

Metric Value What It Means
Total Stories 33 Autonomously planned, implemented, verified, and reviewed
Production Repos 3 Real Solana ecosystem projects, not toy examples
Regressions 0 Verifier + Tester + Reviewer pipeline catches issues before merge
Human Intervention 0 From task description to tested PR -- fully autonomous

Your agents write code.
Antfarm makes them ship it.

The orchestration layer for multi-agent software development. Drop-in agent pipelines that turn feature requests into tested PRs. Proven, not theoretical.