Allen is an open-source AI engineering team that runs on your repos — specialist agents, deterministic orchestration through workflows, and approval gates you control. Drive it from chat, Slack, or Linear.
Most AI coding tools give you one assistant in your editor. Allen is an orchestration framework that runs many — a planner delegates to builders, builders fan out in parallel, a reviewer signs off, all on a workflow you can replay step for step.
A product-manager scopes the work and hands it to the engineering-lead, who spawns specialists — technical-designer, backend- and frontend-developer, test-writer, code-reviewer — each with its own role, model and tool scope.
Long-running work fans out across independent lanes — frontend, backend, tests — each in its own isolated branch and terminal. They run concurrently, then converge.
Long jobs run as graphs with explicit nodes, structured outputs and human checkpoints. Retries on failure, branches on condition. Every run is fully traced — replay it step by step.
Allen plugs into the tools you're on — Linear, GitHub, Slack — and runs on the AI coding agents you trust. Bring your own keys.
You don't need to be at your desk. Allen plugs into your Slack workspace as a teammate — same agents, same workflows, same context as the UI.
Spin up Allen and a full engineering org is already running — six teams, 20+ agents, pre-built workflows. Edit them in chat. Add your own without writing a line of YAML.
Team leads delegate, specialists execute. Use as-is — or edit, extend and add your own.
Plan, build, fix, review, document — ship-ready graphs you can run unchanged or fork.
You don't have to hand-author YAML or click through a builder. Describe the agent you want, or the pipeline you need — Allen scaffolds the prompt, picks the model, scopes the tools, wires the graph, and saves it to your Allen instance. Tweak, replay, ship.
agents/security-reviewer.yml · scoped to read_file,grep,github.review · modelsonnet. Ready to commit?bug-fix-by-severity· new node security-review added before open_pr, gated on approval.The built-in team is a starting point, not a ceiling. Add agents with their own prompts and tool scopes. Compose workflows from a visual canvas or write them as YAML. Both ship in the repo.
Pick a model, write the system prompt, scope which tools and MCP servers it sees, and declare who it can delegate to. Drop it in and Allen picks it up automatically.
name: security-reviewermodel: sonnettools: [read_file, grep, github.review]delegates_to: [test-writer]prompt: | # role Audit diffs for secrets, injection, unsafe deserialization. Block if found. Suggest a fix, never apply.
Compose nodes on a canvas or describe them in chat — Allen drafts the YAML. Conditional branches, parallel lanes, retries, structured outputs, human checkpoints. Versioned in your repo.
Allen ships with an inbuilt context engine. It starts with the context you already have and keeps learning as you work — then feeds each agent only the slice that's relevant. No noise. No wasted tokens. No hallucinated detail.
Claude Code and Codex are great for direct single-agent coding sessions. Allen is for coordinated engineering work: multiple agents, delegated tasks, isolated workspaces, workflow checkpoints, live traces, artifacts, and repeatable execution across real repositories.
No. Allen builds on them. It can run Claude Code CLI and Codex as execution engines, then adds orchestration, visibility, approvals, integrations, and team-level workflow management around them.
Allen keeps agent work observable. Every run has live traces, logs, tool calls, artifacts, isolated workspaces, and human checkpoints, so teams can see what happened, review outputs, and approve important steps before work continues.
Yes. Allen can integrate with Slack so teams can communicate with agents from existing Slack threads, dispatch work, follow progress, and keep engineering conversations connected to the actual agent execution.
Yes. Allen is MIT licensed and self-hosted, so your team can run it on your own infrastructure, connect your own repositories and tools, inspect the code, and adapt the system to your workflow.
Run Allen on your laptop, on a dev box, or in your own cloud — it's the same binary. The setup script installs Database, the Claude Code CLI and Codex CLI, generates keys, and verifies the toolchain end to end.
./scripts/setup.sh — toolchain, secrets and health check, automatic.localhost:5173, or deploy to your infra.