Build AI coworkers by chatting. Share them with your team. Talk to them in Slack, WhatsApp or Telegram.
Agenta is an open-source workspace where you build specialized agents that automate and augment your work.
You build agents by chatting with them. You describe the work, connect the apps they need, and improve them through feedback.
You can work with your agents in the web app, or talk to them where you already are: Slack, Telegram or WhatsApp Business.
For recurring work, you can create automations. Your agents run on a schedule or when an event occurs in a connected app, and you can see what every run did.
Step-by-step videos. Each one builds a real agent from scratch, by chatting.
Build your first AI coworker · 10 min Connect your tools, give the agent a role, save a skill, and automate sales-call prep. |
Track your competitors while you sleep · 7 min Give an agent web search, schedule it for every Monday, and get the digest in Gmail. |
Build an inbound outreach agent with no code · 17 min Research every new lead, qualify it, and send a personalized reply. |
More tutorials on our YouTube channel.
Agenta is the wrong tool if you are looking for:
-
an agentic development environment for writing code, like ocra, Claude Code or OpenCode.
Agenta is built for the work around and beyond code: IT ops, GTM, HR, legal, finance, support...
-
a single-user personal assistant, like openclaw or Hermes.
Agenta is built for teams: agents are shared, multi-user, and run where the team works (e.g. Slack).
-
a low-code workflow builder, like n8n or Zapier.
In Agenta you do not draw workflows. You build autonomous agents that figure out the steps and solve problems on their own.
Agenta is built for work, with professionals in mind from day one. That means:
- Multiplayer by default. Give your team (or customers) access, set roles, and share agents and chat sessions.
- Transparent. You see each agent's configuration, its versions, and every change.
- Controllable, with you in the loop. You set exact permissions per agent: which tools it can use, and whether it can read or write with them. You decide what it does on its own and what goes through you first.
- Secure. Each session runs in its own sandbox, isolated from every other. Model credentials are never seen by the agent.
- Auditable. Every run is traced. Every cost is tracked.
- Yours. Open source, self-hostable, on your models, with your data on your servers.
- For technical and non-technical teammates. You build agents by chatting, so anyone on the team can build them and work with them, not only engineers.
Talk to your agents in Slack, Telegram or WhatsApp. Your team and your clients use agents from the chat apps they already have, with no new tool to learn. The web app works on your phone too.
Use your Claude or ChatGPT subscription. Run agents on the subscription you already pay for, instead of metered API billing.
Use any model. Hosted models through an API, or self-hosted models with Ollama. Pick a cheap model for routine jobs and a strong one where it matters, and see what each agent costs.
Agents that learn. Correct an agent once and it remembers. Agents keep memory, create and refine their own skills, and update their own instructions, so they need less supervision every week.
Automations. Run agents on a schedule or when an event occurs in a connected app. See the history of every run, and continue a run's session to fix or improve it.
Each agent has its own computer. A browser and a file system, so agents can work on any website and any file, not only on apps that have an integration.
Integrations. Connect your agents through MCP, or to more than 1,000 apps through Composio, including Gmail, Slack, Notion, and GitHub. Add your own secrets for the tools that need them.
Workspaces for you and your agents. Work together on files in a shared workspace: documents, research, a wiki, with comments and previews. Agents can also build small apps, like a board or a form, for your team to use.
Teams of agents. Agents can call other agents and hand work to each other, so one coworker can coordinate specialists.
Skills and templates. Start from templates, bring in skills from the agent ecosystem, and grow your own skill library over time.
The fastest way to try Agenta.
Paste this into your agent and it will walk you through setup and testing:
1. Install the Agenta self-hosting skill: npx skills add Agenta-AI/agenta-skills
2. Help me self-host Agenta with its repository.
For more details, read the self-hosting documentation.
These products are designed for building workflows with predefined steps. Agenta is designed for work that requires an agent to plan, use tools, and adapt its approach. Use a workflow builder for predictable processes. Use Agenta when the work needs judgment, or when you want the same agent to work with you in chat and run in the background.
Claude Cowork provides a workspace built around Claude. Agenta is open source and lets you choose your harness and model, inspect the components of your agents, and run them interactively or in the background.
These coding agents provide the execution layer that plans work and uses tools. Agenta adds the shared workspace around that execution layer: files, team access, triggers, versions, and traces. Agenta supports Claude Code, Pi, and Codex today. Support for more harnesses is on the roadmap.
See the complete roadmap. Want to help with one of these items? Open a discussion or contribute.
Agenta is open source under the MIT License. You can inspect the code, run it yourself, and help shape what we build next.
- Read the documentation
- Report a bug
- Request a feature or share an idea
- Read the contributing guide
- Join the Slack community
If Agenta is useful to you, star the repository and tell us what you build.
Consider giving us a star! It helps us grow our community and gets Agenta in front of more developers.
Thanks goes to these wonderful people (emoji key):
