Skip to main content

Changelog

New features, improvements, and fixes in Agenta.

v0.128.0

Know what an agent is missing before you run it

When something an agent uses goes missing, the new app now tells you before a run fails. The item shows Needs setup in the agent editor, with the reason under it. If the model can't run, a banner above the chat box says so, and Send stays off until you set it up.

  • A missing MCP server, secret or connected app no longer stops a run. The agent runs without it.
  • A scheduled or event run whose model needs setup is recorded once as skipped, instead of being retried five times.
  • Invite links open in the new app, and new teammates land in the project they were invited to.
Read more
v0.127.1

Less repeated work when agent runs start and finish

Release concept: Less repeated setup work when agent runs start and finish.

The runner now avoids several repeated steps when starting an agent session. It reuses local tool binaries and Pi skill copies where applicable, mounts drives in parallel, and reuses a harness capability probe.

When its trace is ready, a completed Pi turn no longer waits for the full trace-drain polling window. Organization owners and admins can test their connections from a project that does not use them. Conversation replay no longer cuts through a character and leaves a chat unable to continue.

Read more
v0.127.0

First-agent onboarding and more dependable event automations

Release concept: A path guides a new account toward its first agent.

New accounts now enter first-agent onboarding when their project has no agents. The questions use the same role, company-size and referral choices as the signup survey, so people do not need to answer conflicting versions of them.

  • Event automations now receive Composio V3 webhook events instead of silently dropping them.
  • Editing an automation's event filters now updates its Composio trigger.
  • Organizations replace workspace records; self-hosted operators should read the upgrade notes before deploying.
Read more
v0.126.0

Use AI provider connections across an organization

Organization owners and admins can now add an AI provider connection once for the organization instead of saving the same credential in every project. In Settings, the Organization tab shows the connections available to the current project.

  • Choose which projects may use a connection; selecting individual projects requires the supported Enterprise entitlement.
  • Pick an organization connection explicitly in an agent's model picker.
  • Agent runs using a saved bare model ID now mark the matching model row as selected.
Read more
v0.125.2

Continue channel chats in Agenta

You can now continue a Slack or Telegram chat in Agenta. Members who can send to a project-visible channel chat can write in the app, and the agent's answer also goes back to the channel thread.

  • A private agent can stay connected to a channel through a connection its owner made.
  • A new channel chat with that private agent belongs to its owner and starts private.
  • Fixes keep first messages and more of a long turn's work when a chat changes or a turn ends.
Read more
v0.125.3

Security updates and agent-run fixes

This release updates vulnerable dependencies across the web app, API, SDK and runner.

On self-hosted setups that override HOME, Codex runs no longer wait indefinitely on their first turn. Agent runs send the correct project even without an authorization header. The model picker shows a readable Kimi K3.1 Pro label, and session names no longer show a lock icon.

Read more
v0.125.1

More reliable built-in Gemini runs

Built-in Gemini agent runs now use Google's native protocol instead of its OpenAI-compatible endpoint. This changes the route used for model calls that could stall on follow-up turns.

Bedrock and SageMaker credentials now travel with each call instead of being set in the shared process environment. The Python SDK also accepts an AGENTA_API_URL without a URL scheme, and CSV preview comments no longer show a stray colon when a column has no header.

Read more
v0.125.0

Private agents and chats, and sharing with your project


New chats and agents that you create in Agenta now start private. Only you see them until you share them. You can share a chat with chosen people as Can view or Can send, or make it visible to the whole project. You can also make an agent private.

Chats and agents that existed before this release keep working as before. They are visible to the project, with their creator as owner.

3 improvements and 5 fixes in this releaseRead more
v0.124.0

Project access, permissions and runs that act as their creator

Agenta now decides project access in one place. Creating, renaming and deleting a project follows organization roles. An API key acts only in its own project for management actions. When you remove a member or delete a key, the change takes effect at the next request.

Automations, evaluations and approvals run as the user or key that created them. If that identity loses access, the run stops and tells you why.

1 improvement and 1 fix in this releaseRead more
v0.124.1

Projects use only their own provider keys

Prompts and evaluators now run on the provider keys saved in the project. A project without a key for a provider gets the error "No API key found for model". Add the key in Settings > Providers & Models.

Custom providers need their own API key, except Bedrock, SageMaker and Vertex. Self-hosted Enterprise deployments can opt in to keys set on the services container.

1 improvement and 2 fixes in this releaseRead more