Alook AI Workforce Layer
Coordinate local coding agents as a traceable team with roles, email, tasks, and schedules.
- Source repo
- alookai/alook
- Stars
- ★ 1.2k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 42/100 · Major gaps
At a glance
- Works with
- Universal · cross-platformCodex · Claude Code
- You'll need
- Typical use
- A solo developer wants Codex, Claude Code, or OpenCode to collaborate around one codebase under development, operations, and research roles.
- Main limitation
- The supplied documentation does not provide complete prerequisites, runtime versions, or credential setup beyond
npx @alook/app onboard. - Source review
- 42/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Alook is an open-source, self-hosted platform for turning local AI coding agents into a collaborative workforce. It assigns agents roles and email addresses, with an org chart, kanban board, calendar, and dashboards for coordination. Agents run on the user's machine with access to local tools and codebases, while Alook connects them to email, calendars, and external services. Its architecture diagram names `@alook/cli`, an agent workdir, `@alook/app`, Email, WebSocket, SQLite, and file storage. The repository says it is built with Next.js, Cloudflare Workers, and Bun, and lists Claude Code, Codex, and OpenCode as available agents.
Running npx @alook/app onboard starts a guided setup that connects the machine, detects runtimes, and deploys a first agent company; the stated local endpoint is http://localhost:15210. Users define an org chart and assign roles such as dev, ops, and research, while agents communicate through individual email addresses. The kanban workflow assigns work and tracks progress; the README says agents pick up work, update status, and close issues. The calendar supports recurring tasks, reminders, and daily routines. The architecture diagram shows @alook/cli polling from an agent machine to a hosted machine containing @alook/app, Email, and WebSocket, which interact with SQLite and file storage.
- A solo developer wants Codex, Claude Code, or OpenCode to collaborate around one codebase under development, operations, and research roles.
- An open-source maintainer needs a kanban workflow where agents can take assigned work, update its status, and close completed issues.
- A small team wants separate addresses for local coding agents so human-to-agent and agent-to-agent communication remains in one inbox-based system.
- An operator wants to schedule daily routines, recurring work, and reminders for a set of agents through a calendar.
- A technical lead needs recorded instructions, decisions, and replies to review how agent work was carried out.
How do you install or deploy this agent?
Run npx @alook/app onboard. The documented flow connects your machine, detects runtimes, and deploys your first agent company; then open http://localhost:15210. The supplied documentation does not specify prerequisite runtime versions beyond use of npx, required credentials, or the detailed email-claim configuration. It only says unique @alook.ai agent addresses can also be claimed through alook.ai.
How do you use this agent?
Run npx @alook/app onboard, then open http://localhost:15210. Define the org chart, assign agent roles, and place work on the kanban board; use agent email for communication and the calendar for recurring tasks, reminders, and routines. Optionally, claim unique @alook.ai addresses for agents through alook.ai.
What are this agent's strengths and limitations?
- It explicitly supports Claude Code, Codex, and OpenCode, making it a coordination layer for existing local coding-agent workflows.
- It combines org structure, per-agent email, kanban work tracking, and calendar scheduling rather than exposing only a task launcher.
- Agents stay on the user's machine with local tool and codebase access, while the README also describes recorded instructions, decisions, and replies for traceability.
- The supplied documentation does not provide complete prerequisites, runtime versions, or credential setup beyond
npx @alook/app onboard. - Agents have full access to local tools and the codebase; no permission isolation, approval workflow, or least-privilege controls are documented.
- Although remote reachability is claimed, the documentation does not describe authentication, network exposure, or the security boundary for that access.
- Cursor, Hermes, and OpenClaw are marked “Coming Soon,” so they are not current supported integrations.
How does this agent compare with similar options?
Alook positions itself as an orchestration layer around Claude Code, Codex, and OpenCode rather than a replacement for those coding agents; it adds roles, inboxes, and an always-on runtime. The supplied documentation does not compare their features, performance, or cost.
Key facts side by side with the most closely related agents.
| Agent | Source review | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|
| Alook AI Workforce Layer This agent | 42 · Major gaps | ★ 1.2k | today | TypeScript | Codex · Claude Code |
| COMPASS Skills | 85 · Good | ★ 735 | 28d ago | Python | Codex · Claude Code |
| OpenHuman | 51 · Major gaps | ★ 40k | today | Rust | — |
| Memmy — Personal AI Agent & Local Memory Hub | 42 · Major gaps | ★ 2k | today | TypeScript | Codex · Claude Code · OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: README claims local-first and codebase never leaves, but no specific permission controls or least-privilege mechanisms are provided; CLI onboard flow may involve user confirmation but not detailed; data flow diagram shows client-cloud communication but data flow and storage details unclear; sensitive data handling (e.g., encryption keys) mentioned in package.json but not detailed; dependency security not mentioned; external effects (e.g., sending emails, deployment) exist but user confirmation not specified; rollback mechanism not mentioned; source attribution unclear. Deductions: lack of implementation details and evidence.
Evidence: Project structure clear, multiple modules (web, cli, email-worker, etc.) consistently named, tests cover some functionality; dependency management uses pnpm but no lockfile or availability statement; failure messages shown in tests (e.g., migration failure exits), but overall error handling not fully demonstrated. Deductions: dependency availability unverified, failure message coverage limited.
Evidence: README clearly defines target users (developers, ops, researchers) and scenarios (collaboration, email, kanban, calendar); capability boundaries not clearly stated (e.g., limited agent types); trigger precision not detailed (e.g., task assignment mechanism); environment fit good (local, cloud, desktop). Deductions: capability boundaries and trigger precision lack detail.
Evidence: Information architecture clear (README, CONTRIBUTING, LICENSE); install instructions simple (npx command) but no detailed configuration; naming stable (consistent module names); examples and FAQ lacking; known limitations not mentioned; license Apache-2.0; versioning has auto-tag and changelog generation but no CHANGELOG file; maintenance responsibility unclear. Deductions: examples, FAQ, known limitations missing, versioning evidence insufficient.
Evidence: Output usability not specified (e.g., report formats); marginal value high (provides collaboration platform); cost-benefit not assessed (e.g., resource consumption). Deductions: output usability and cost-benefit lack evidence.
Evidence: README feature claims lack specific implementation links; cross-source verification insufficient (only README and some tests); facts and inferences not clearly separated. Deductions: lack of traceability and cross-validation.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- No specific implementation of least privilege or permission controls; agents may have excessive permissions.
- Data flow and storage details are opaque; sensitive data handling not fully described.
- Dependency security not mentioned; supply chain risk exists.
- Rollback mechanism missing; operations may not be recoverable after failure.
- Known limitations not stated; users may misunderstand feature boundaries.
FAQ
Which agents are currently available?
Can agents access my local code?
Do agents need email addresses?
@alook.ai addresses can be claimed through alook.ai, but does not document that setup in detail.