agentes.chat: Open-Source Slack Alternative with AI Agents

What changed

agentes.chat has released its open-source, self-hosted platform, positioning itself as an alternative to Slack where human teams and AI agents can collaborate within channels and direct messages. The system allows AI agents to join workspaces with defined roles, budgets, and owners, operating with a degree of autonomy while maintaining safety through approval mechanisms for sensitive actions. Agents can leverage various large language models, including Anthropic's Claude, OpenAI's ChatGPT/Codex, Google's Gemini, and xAI's Grok, with support for any model available via OpenRouter. The platform emphasizes agent capabilities such as having their own dedicated computer and browser for executing tasks, managing routines on schedules or event triggers, and maintaining memory of past interactions and decisions. It also includes features for approvals on sensitive operations like merges or deployments, and integrates with external services via MCP tools, supporting a wide range of applications through Composio. The latest release, desktop-v0.7.5, indicates ongoing development and feature refinement.

Why it matters for builders

Builders can leverage agentes.chat to create sophisticated AI-powered collaboration environments. The platform's open-source nature and self-hosting capabilities provide control and customization, allowing developers to integrate their own LLM keys and tailor agent behaviors. The architecture supports agents that can perform complex tasks, interact with external systems, and learn over time, offering a foundation for building advanced AI assistants that function as true team members rather than isolated bots. The inclusion of features like agent-specific browsers and computers, along with a robust approval system, addresses key challenges in agent reliability and safety.

Practical impact

Developers can deploy agentes.chat using Docker Compose with PostgreSQL as the primary dependency, or opt for self-hosting without Docker for simpler setups. The platform supports web, desktop (macOS, Windows, Linux), and phone applications, ensuring accessibility across devices. Integration with various LLMs is managed through API keys configured within the application. For developers looking to extend agent capabilities, the platform supports custom skills via Markdown files and integrates with a wide array of third-party services through MCP connectors. The ability to import agents from other platforms like Grok Bot and ChatGPT dots also facilitates migration and experimentation. The project is licensed under AGPL-3.0, with specific permissions for app store distribution and attribution.

Caveats and source limits

The source material indicates that the agent browser functionality is currently in beta. Similarly, the Postgres connector for querying databases and agent mailboxes for reading/sending emails are also in beta. The desktop and phone applications are also noted as being in beta. While the platform supports a wide range of LLMs, the specific performance or cost implications of using different models are not detailed. The source does not provide independent benchmark results for agent performance or system scalability. The AGPL-3.0 license requires modified versions to retain attribution, which is a key consideration for commercial use.