CowAgent
A self-hosted, multi-channel assistant that plans tasks and uses tools, skills, memory, and knowledge to carry them out.
- Source repo
- zhayujie/CowAgent
- Stars
- ★ 48k
- Last updated
- 1d ago
- License
- MIT
- Primary language
- Python
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- An individual who wants a 24/7 assistant on a personal machine or server can use the Web Console to chat, configure models, and connect channels.
- Not a fit if
- Users unwilling to give an agent local OS access
- Cost-sensitive users who only need plain chat
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
CowAgent is an open-source super AI assistant and Agent Harness intended to run continuously on a personal computer or server. Messages enter through the Web Console or supported messaging channels; the Agent Core plans and executes with models, memory, knowledge, tools, and Skills before returning results to the originating channel. It includes a three-tier long-term memory design, an automatically curated Markdown knowledge base, and visual knowledge-graph browsing. Its tool layer covers file operations, bash, browser automation, scheduling, web search, and MCP integration, while Skills package higher-level workflows. It can be deployed locally, with Docker, or on a server, with configuration managed through the Web Console.
CowAgent receives text, images, voice, and files from the Web Console, Telegram, Slack, Discord, WeChat, Feishu/Lark, DingTalk, and other listed channels. Its Agent Core decomposes complex goals and repeatedly calls built-in tools including read, write, edit, ls, bash, send, memory, env_config, web_fetch, scheduler, web_search, vision, and browser. It can also load MCP servers from mcp.json over stdio or SSE. It stores information across conversation context, daily memory, and MEMORY.md, and curates structured information into a Markdown wiki. Users manage Skills with /skill list, /skill search <keyword>, and /skill install <name>, while responses or files are delivered through the initiating channel.
- An individual who wants a 24/7 assistant on a personal machine or server can use the Web Console to chat, configure models, and connect channels.
- A team using Telegram, Slack, Discord, or enterprise messaging can run one Agent instance across several channels in parallel.
- A user who needs an assistant to read, edit, and send local files while running terminal commands can include file I/O and
bashin a multi-step task. - A personal knowledge-management user can turn everyday conversations into retrievable memory and a topic-organized Markdown knowledge base.
- A developer with existing MCP servers can add capabilities through
mcp.jsonusing stdio or SSE without writing integration code.
How do you install or deploy this agent?
Linux/macOS: bash <(curl -fsSL https://cdn.link-ai.tech/code/cow/run.sh); Windows PowerShell: irm https://cdn.link-ai.tech/code/cow/run.ps1 | iex; Docker: run curl -O https://cdn.link-ai.tech/code/cow/docker-compose.yml, then docker compose up -d. After startup, open http://localhost:9899, configure a chosen model provider in the Web Console, and begin using the agent. The supplied material does not specify the exact credential fields or acquisition process for each provider. For server access, set web_host to 0.0.0.0 in config.json, set web_password, and open port 9899.
How do you use this agent?
Open http://localhost:9899, then use the Web Console to configure models, connect channels, install Skills, and chat with the Agent. After installation, manage the service with cow start, cow stop, cow restart, cow status, cow logs, and cow update; use cow skill install <name> to install a Skill or cow install-browser to install browser automation. In chat, use /skill list, /skill search <keyword>, and /skill install <name> to manage Skills.
What are this agent's strengths and limitations?
- Combines task planning, iterative tool use, long-term memory, a knowledge base, and multi-channel messaging in one self-hosted Agent Harness.
- Its three-tier memory model—context, daily memory, and
MEMORY.md—is paired with Deep Dream consolidation, an auto-maintained Markdown knowledge base, and knowledge-graph browsing. - It provides built-in file, terminal, browser, scheduler, and search capabilities while supporting MCP stdio/SSE integration and hot reload through
mcp.json. - Chat, vision, image generation, ASR, TTS, and embeddings can be routed to different providers, with multiple named model providers supported.
- The Agent can access the local operating system; the project explicitly advises deployment only in trusted environments, so adopters must assess permissions and isolation.
- Agent mode consumes substantially more tokens than regular chat, making operating cost dependent on model choice and usage.
- Public server access requires manual
config.jsonchanges, aweb_password, and firewall or security-group configuration. - Although many model providers are listed, the supplied material does not specify provider-by-provider credential setup, quotas, or feature limitations.
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 |
|---|---|---|---|---|---|---|
| CowAgent This agent | 52 · Major gaps | CLIFree + model costs | ★ 48k | 1d ago | Python | OpenAI API · Claude API |
| Octop Self-Hosted Assistant | 67 · Some gaps | CLIFree + model costs | ★ 12k | 1d ago | Python | Codex · Claude Code · OpenAI API |
| OpenAgent | 27 · Major gaps | CLIFree + model costs | ★ 5.7k | 4d ago | Go | OpenAI API · Claude API |
| Engramory Memory | 63 · Some gaps | Agent plugin / skillFree | ★ 201 | 3d ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows web console password protection and security hardening, but lacks detail on least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, and source attribution. Thus each criterion scores 1 due to partial evidence.
Self-consistency scores 2 as README, tests, and code structure align; dependency availability scores 1 as dependency list exists but not verified; failure messages scores 1 as tests cover some error handling but not comprehensive.
Audience and scenarios score 2 as clearly targeting individuals and enterprises; capability boundaries score 2 as tools and skills are listed; trigger precision scores 2 as CLI and skill commands exist; environment fit scores 2 as supports multiple platforms and models.
Information architecture scores 2 as clear documentation structure; install notes score 2 as one-line installer and Docker provided; naming stability scores 2 as renaming notice exists; examples and FAQ score 2 as quick start provided; known limitations score 2 as disclaimer exists; license scores 2 as MIT; versioning and changelog score 2 as detailed changelog; maintenance responsibility scores 2 as contribution guide exists.
Output usability scores 2 as web console and CLI exist; marginal value scores 2 as multi-channel and skills provided; cost-benefit scores 2 as cost tips and free open source.
Claim traceability scores 1 as doc links exist but not verified; cross-source corroboration scores 1 as community links exist but not verified; fact-inference separation scores 1 as disclaimer exists but not explicit.
- Static review cannot verify actual runtime behavior; all security claims need further testing.
- Dependencies are not pinned, posing supply chain risks.
- No explicit least-privilege strategy; Agent may have excessive system access.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
🌐 [English](https://docs.cowagent.ai/releases/v2.2.2) | [中文](https://docs.cowagent.ai/zh/releases/v2.2.2) 💬 Guest Pages Share an Agent as a standalone chat page, or embed it in your website as a chat window. Visitors open the link and start chatting without signing in, which suits product inquiries, knowledge base Q&A, customer support, and similar scenarios: - Standalone pages: each page gets its own link, visitors keep their chat history, and the name, icon, theme color, welcome message, and suggested questions are all customizable. <Frame> …View on GitHub ↗ -
Release notes excerpt
🌐 [English](https://docs.cowagent.ai/releases/v2.2.1) | [中文](https://docs.cowagent.ai/zh/releases/v2.2.1) 📁 Artifacts A new Artifacts page brings together the files created in your conversations, in both the Web console and the desktop client: - Automatic collection: documents, web pages, images, and videos your Agents create show up in Artifacts automatically, on one timeline across all Agents. - Preview and trace back: filter by type or Agent, search, preview in place, and jump back to the conversation that produced a file (Thanks @rudycelekli #3732). …View on GitHub ↗ -
Release notes excerpt
🌐 [English](https://docs.cowagent.ai/releases/v2.2.0) | [中文](https://docs.cowagent.ai/zh/releases/v2.2.0) 🧩 Capabilities Center The former Skills menu is now Capabilities, a single place in the Web console and desktop client to manage built-in tools, MCP tools, and skills: - Skill installation: install from Cow Skill Hub, GitHub, or ClawHub, or upload an archive, a SKILL.md, or a folder. Every skill is previewed first and only installed after you confirm. …View on GitHub ↗
FAQ
Can it run without a managed cloud runtime?
What permissions does the assistant need?
bash; the disclaimer states that the Agent can access the local operating system and should be deployed only in trusted environments.Can it use existing MCP servers?
mcp.json can integrate MCP servers; the material explicitly supports stdio and SSE transports and describes hot reload.