Automatos AI Work OS
Run a coordinated team of AI agents on infrastructure you control, with shared tools, knowledge, and workflows.
- Source repo
- AutomatosAI/automatos-ai
- Stars
- ★ 71
- Last updated
- today
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 54/100 · Blocked
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude APIClaude Code (Partial support)
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A small team that wants to delegate research, writing, and analysis on its own infrastructure can use Auto and specialist agents, then track results on the board.
- Not a fit if
- Teams seeking an account-based SaaS they can use immediately
- Users unwilling to configure Docker and model-provider credentials
- Source review
- 54/100 · Blocked
What does this agent do, and when should you use it?
Automatos combines an Auto chat interface, specialist Agents, Playbooks, and Missions in a self-hostable AI work platform. Auto routes requests to an agent, tool, or workflow; agents can use their own model routes, skills, and tools, while a board, calendar, and reports track work. The platform connects to the Composio catalogue and multiple model providers, and indexes documents for RAG and a Knowledge Graph. It produces reports, code, content, analysis, and other Deliverables; the local edition runs with Docker Compose and keeps agent data and work directories on infrastructure under the operator’s control. The repository also describes a hosted Studio edition, with beta planned for November 2026.
A user submits a request through Auto chat, and the Universal Router routes it to an agent or workflow using rules, semantic similarity, or an LLM. Before building the prompt, the platform can rank tool actions semantically, then call configured Composio integrations or native platform tools; Playbooks run reusable multi-step automations, and Missions coordinate agents. Users can upload documents or sync cloud files; the platform chunks, embeds, and indexes them for RAG, the Knowledge Graph, CodeGraph, and the memory layer. Tasks move through the Command Center board and are executed by agents, with reports, code, content, analysis, or files delivered as outputs. Optional local Session mode lets a paired host run the user’s own Claude Code, Codex, or GitHub Copilot CLI sessions inside registered directories.
- A small team that wants to delegate research, writing, and analysis on its own infrastructure can use Auto and specialist agents, then track results on the board.
- An operations team with recurring multi-step work can create Playbooks, configure triggers or schedules, and coordinate agents with Missions.
- A team that needs agents to search internal material can upload documents or sync files for retrieval through RAG and the Knowledge Graph.
- A developer working in Git repositories who wants Claude Code to run locally can enable Session mode, route tickets into worktrees, and collect Deliverables.
- A team that needs services such as GitHub, Slack, Jira, or Shopify can provide its own Composio API key and assign integrations to selected agents.
How do you install or deploy this agent?
The local edition requires Docker Engine 25 or later and Docker Compose 2.24.4 or later. Clone the repository and start the stack:
git clone --depth 1 https://github.com/AutomatosAI/automatos-ai.git
cd automatos-ai
make upmake up generates three random secrets, pulls about 1.3 GB of images, and starts the services. Open http://localhost:3000; API documentation is at http://localhost:8000/docs. The local edition has no login. To use a model, provide a key for a provider such as OpenRouter, OpenAI, Anthropic, DeepSeek, or NVIDIA; keys can also be added later under Settings → API Keys. Composio integrations require your own COMPOSIO_API_KEY.
How do you use this agent?
After startup, install a model route under Marketplace → LLMs and add the matching provider key; alternatively, add keys under Settings → API Keys. Submit a request in Auto chat or run the seeded Two-minute brief Playbook. To enable local Claude Code session agents, set the following in .env:
CLI_RUNTIME_ENABLED=true
LOCAL_PROJECTS_DIR=/path/to/your/projectsRun make up again, then open Settings → Session mode to get a pairing code and run the displayed pairing command on the machine that holds the project directories. Deliverables are written to the directory configured by AUTOMATOS_WORKSPACE_DIR.
What are this agent's strengths and limitations?
- The local stack includes Postgres, pgvector, Redis, and MinIO and can run on infrastructure the operator controls.
- The provider registry documents routes for OpenRouter, OpenAI, Anthropic, Google, DeepSeek, NVIDIA, Azure OpenAI, AWS Bedrock, and Grok.
- Agents, Playbooks, Missions, the task board, and calendar work together in one platform to track execution and reports.
- Session mode can use the user’s own Claude Code session as an agent and bring its outputs, session log, and permission questions into the task flow.
- The initial deployment requires Docker Engine and Docker Compose and downloads about 1.3 GB of images.
- The local edition requires model-provider credentials; the Composio catalogue also needs a separate
COMPOSIO_API_KEY. - The local edition is a single-workspace, single-operator setup; multi-team workspaces and accounts belong to the hosted edition.
- Session mode requires a paired host on the machine with the repositories and is available only in the local edition.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Automatos AI Work OS This agent | 54 · Blocked | Self-hosted serviceFreemium | ★ 71 | today | Python | OpenAI API · Claude API |
| AgenticOS Team Agent Workspace | 68 · Some gaps | Self-hosted serviceFree + model costs | ★ 118 | today | Python | OpenAI API · Claude API |
| CrewAI Studio | 51 · Major gaps | Web appFree + model costs | ★ 1.4k | 2mo ago | Python | OpenAI API · Claude API |
| EDDI Conversational AI Orchestration | 68 · Some gaps | Self-hosted serviceFree + model costs | ★ 387 | 3d ago | Java | OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
The documentation describes session agents as restricted to registered directories, using a worktree per ticket and never pushing; it also says session permission questions surface as approval cards and that starting a board task is an approval action. These support user control in ordinary workflows. Deductions: PRD-70 describes a skills-import path without an admin check or approval and reports branch-parameter git argument injection that could execute arbitrary commands on the backend; its proposed fixes remain a plan. API keys, Composio, model providers, and knowledge sync also create external data flows that are only partly explained. Unverified publisher identity is not itself a deduction.
The files identify Docker, database, Redis, and model-routing dependencies, and state that native tools remain available without a Composio key. Deductions: PRD-44, PRD-70, and PRD-156 give differing or conflicting accounts of security status and completion, while PRD-70 lists serious fixes as pending. User-facing failure handling is only described in fragments, so stronger scores are not supported.
The README covers local and hosted editions, API and session agents, multiple model providers, agent teams, Playbooks, knowledge bases, and several deployment environments. It also states that the local edition has no login and uses one workspace and operator. Deductions: coverage of broader scenarios, capability boundaries, and trigger behavior is incomplete, and some routing and automation claims are presented primarily as product descriptions.
The README is organized and links to Quick Start, SETUP, QUICKSTART, self-hosting, documentation, and support channels; Apache-2.0 is clearly identified. Deductions: the supplied material does not establish a clear release-version or regular changelog practice, and the statement that the project is built by the Automatos AI team does not specify maintenance responsibility or update commitments. Conflicting security-document statuses also weaken documentation stability.
The product combines reusable agents, tools, Playbooks, a task board, knowledge retrieval, deliverables, and cost analytics, which could support useful workflows. The README gives setup requirements and a first example Playbook. Deductions: these are mostly documented feature and value claims, without execution evidence for output quality or realized benefit; local deployment also entails substantial image-download and service costs.
The README provides configuration names, startup commands, component details, and documentation links. Security materials identify file paths, describe findings, and list remediation and test plans, making some claims traceable. Deductions: the security materials conflict on remediation status, and the governance security page says its webhook discussion has no direct code citations; several product and security claims cannot be corroborated across the supplied materials.
- PRD-70 reports branch-parameter injection in skills import and other git-clone paths, with potential arbitrary command execution on the backend. The supplied remediation is a pending plan, leaving a documented severe risk unresolved.
- The security documents conflict on whether findings have been fixed, so the current code state cannot be confirmed from these materials. No code or tests were executed for this assessment.