Agenta
Agenta is an open-source workspace for building and using AI agents with shared context, files and connected tools. Its product page presents templates for work such as code review and describes interactive tasks that can later run on a schedule. It offers both a hosted service and a self-hosted deployment.
Configuring an agent's work
An agent can combine instructions, reusable skills and tools connected through MCP. The workspace includes a browser and files, with shared agents and conversations described in the plan comparison. Start with a bounded task and the tools it needs before giving it a wider set of integrations.
The publisher describes permissions that decide which actions can run without intervention and which require approval. Its roadmap lists batch tool approvals and persistent agent sessions among shipped work. Use approval boundaries for consequential actions, and inspect the result before treating an agent's completed task as accepted work.
Schedules, traces and roadmap limits
Agenta supports interactive runs, schedules and event triggers. The pricing FAQ defines a run as a tracked execution that creates a trace: a message, annotation or scheduled or event invocation counts as a separate run. Traces record model calls, tool calls, timing, outputs and errors.
That record helps a team investigate a failed or unexpected action. Trace retention is separate from retention of chats, project files and persistent context, so do not equate a short trace window with deletion of all workspace material.
The homepage promotes Slack and Telegram conversations, but the captured roadmap places messaging channels under building next. Verify channel availability in the current product rather than assuming it is already included. The roadmap also distinguishes planned skill-registry and gateway work from shipped features.
Hosted usage and self-management
The pricing page separates cloud plans from self-hosted options. Hosted tiers differ in included runs, team access, trace retention and administration. The free cloud tier has a team-member limit; paid cloud plans describe usage-based runs.
Model costs are not included: users connect their own provider credentials and pay the provider. Self-hosted users also operate the application, storage and runner themselves. The MIT-licensed edition includes the core workspace, while commercial support and additional enterprise features have separate terms.
Evaluate a real task and inspect its trace before estimating recurring usage. An unlimited agent count does not make provider tokens, hosting or additional cloud runs free.