Agenta: Build, run and improve AI agents
Agenta is an open-source workspace where teams build AI agents through chat, run them on schedules or events, and improve them with feedback and evaluations.
What is Agenta?
Agenta is an open-source workspace for building and running AI agents. Agents are created and used through chat, with instructions, reusable skills, tools, integrations and MCP servers attached. Runs can be started interactively, on a schedule, or by an event in a connected tool, and each run produces a trace covering model calls, tool calls, timing, outputs and errors. Agents can be shared with a team, and conversations and files give them persistent context. Agenta can be used as a managed cloud service or self-hosted on your own infrastructure under an MIT licence, with Docker Compose and Helm charts. Cloud plans are billed by agent runs rather than seats: Hobby is free with two team members and 5,000 runs per month, Pro is $29/month with 10,000 runs, and Business is $299/month with governance features such as role-based access control, SSO and a SOC 2 Type II report. Model costs are not included; you connect your own provider credentials.
Who is Agenta for?
- Engineering and platform teams running agents in production
- Teams that need to self-host agents and keep data on their own infrastructure
- Operations, marketing and support teams that want agents working in Slack or Telegram
How much does Agenta cost?
Agenta starts at $29/mo. Always check the vendor’s own pricing page before you buy.
| Plan | Price | What you get |
|---|---|---|
| Hobby | $0 | 2 team members, 5,000 agent runs/month, 1-week trace retention |
| Pro | $29/month | Unlimited team members, 10,000 agent runs/month, $5 per additional 10,000 runs |
| Business | $299/month | 10,000 agent runs/month, RBAC, SSO, SOC 2 Type II report, 3-month trace retention |
| Enterprise | Custom | Custom usage and trace retention, audit logs, custom domains, dedicated support |
| Open source | Free | Self-hosted, unlimited users and runs, bring your own models |
| Enterprise (self-hosted) | Custom | Audit logs, custom domains, deployment support, security reviews |
Is Agenta worth it?
A practical option for teams that want to build, run and inspect AI agents without giving up self-hosting or paying per seat
Where Agenta is strong
- Open-source and self-hostable under an MIT licence, with Docker Compose and Helm charts
- Usage-based cloud pricing with unlimited users, agents and workflows on paid plans
- Tracing, evaluations, version history and a playground for improving agents
Where it falls short
- AI model costs are not included and are paid directly to the model provider
- The free Hobby plan is limited to two team members and 5,000 agent runs per month
How do you get started with Agenta?
- 01Start a free Hobby workspace on Agenta Cloud or self-host the open-source version with Docker Compose
- 02Connect a model provider and the tools or MCP servers the agent needs
- 03Build the agent through chat, then add a schedule or event trigger and review its traces
Agenta: frequently asked questions
What counts as an agent run?
Any tracked execution that creates a trace counts as one run: sending a message to an agent, creating an annotation, or starting an agent from a schedule or event. Each new message, annotation or trigger is a separate run.
Are AI model costs included in the plan price?
No. You connect your own model-provider credentials and pay the provider directly. Supported providers include OpenAI, Anthropic, Gemini, Mistral, Groq, MiniMax, Together AI, OpenRouter, Azure OpenAI, AWS Bedrock and Google Vertex AI, plus OpenAI-compatible endpoints.
What happens when I reach my monthly run limit?
Hobby stops running agents after 5,000 runs in a monthly billing period. Pro and Business include 10,000 runs per month, and additional runs are billed at $5 per 10,000 with agents continuing without interruption. Unused runs do not roll over.
What is the difference between Agenta Cloud and self-hosted Agenta?
Agenta Cloud is operated by Agenta, which runs the application, storage, agent runner, upgrades and backups. With self-hosted Agenta you deploy it on your own infrastructure and manage trace storage and retention yourself.
Agenta alternatives
The closest tools in our directory, matched on category, focus areas and audience.
Teams deploying multilingual AI agents to messaging and voice channels
| Tool | Best for | Free plan | Starting price | Focus areas |
|---|---|---|---|---|
| Agenta | Teams building and operating AI agents they can self-host | Yes | $29/mo | — |
| Gooey AI | Teams deploying multilingual AI agents to messaging and voice channels | Yes | $25/mo | — |
| Rasa | Teams building and governing customer-facing AI agents in-house | Yes | — | — |
| AgentX | No-code platform to build, tune, deploy, and integrate AI Agent Chatbots in minutes. | Yes | — | — |
| Potpie | Large engineering teams automating debugging, testing and delivery | No | — | — |
| RapidCanvas | Large enterprises building custom AI apps | No | — | — |
| OpenServ | Developers and enterprises adding auditable reasoning to agents | Unknown | — | — |
| Promethia | Research and analysis work that needs multiple coordinated agents | Unknown | — | — |
| Rierino | Enterprise teams building AI agents, backends and composable commerce | Unknown | — | — |
| zgi.ai | Teams building governed, self-hosted AI agents on company data | Unknown | — | — |
| 扣子 | Office teams automating writing, slides, tables and media tasks | Unknown | — | — |

