Commander v0.3.0

What's Changed

Prepare intelligent agents for distributed missions with shared credentials,
interactive questions, execution progress and operator-controlled completion.

Missions and execution

• Create and edit mission objectives, actions, rules and schemas through Admin,
  HTTP API, MCP or Admin CLI. Names stay unchanged; running operations retain
  their original mission definition.
• Set completion.decision_by to operator to require explicit completion.
  Model submissions save completion_draft; operators review or edit it in
  Admin before completing through the existing API. Omitted/model policy keeps
  immediate completion.
• Actions can ask durable questions with action.Question. Operators answer
  first; the model may answer after fallback_at. Passing the waiting deadline
  never supplies an answer or completion approval.
• Record execution progress with action.Log, query it through HTTP/MCP and
  follow it in Admin, filtering by task, agent or action.
• Reconcile late execution outcomes after interrupted waits. The default initial
  task-result wait is now 15m, configurable with TASK_TIMEOUT; a timeout does
  not prove the underlying work stopped.
• Inspect and manage agent-owned runtime resources in Admin's Sessions page.

Secrets and deployment

• Save a credential once by name for all requesting agents, including agents
  started later. The controller decrypts requested values for delivery; agents
  and spawner no longer need SECRETS_ENCRYPTION_KEY.
• Playground shares the selected provider key through the API under its canonical
  environment-variable name, preserving any existing saved value.
• Spawner can launch Kubernetes Pods using in-cluster authentication and configured
  namespace, ServiceAccount and child configuration references.
• API, MCP and Admin UI honor platform PORT, with component-specific listener
  settings and explicit flags taking precedence.

Playground and Admin

• Playground retains settings, chat history and locally encrypted provider keys
  in ~/.commander, including remote sessions. --temporary opts out; Docker
  sessions retain state through their mounted volume.
• Provider and connection status stay fixed above the scrolling conversation.
  Completed tool calls replace their pending row, and the operation summary shows
  up to three recent operations.
• Admin opens on Operations, provides shared Secrets in the sidebar and exposes
  connection health through the current connection's Status page.

Commander spawner agent v0.3.0

What's Changed

• Launch native processes, Docker containers or Kubernetes Pods and return each
  runtime's identity through runtime_details.
• Use mounted in-cluster credentials for Kubernetes; child Pods use configured
  references without receiving the secrets encryption key.
• Install github.com/defos-ai packages directly. Other owners trigger an
  execution question: operator first, model fallback after the waiting window.
  Approval applies to that execution; no owner trust is saved.
• Inspect execution logs for package selection, preparation, launch reconciliation
  and registration waits.

Commander builder agent v0.3.0

What's Changed

• Use the shared managed OPENAI_API_KEY delivered by the controller.
• Follow checkout, generation, validation, publication and release stages through
  execution logs, including periodic progress during long stages.
• Run as an optional Compose service or install the package separately. All-in-one
  retains the tooling needed by installed builders.

Commander v0.2.0

Commander v0.2.0

Commander can now build agents, install published packages on demand and create
missions through Admin, MCP or the HTTP API.

New

• Builder agent: generate agents with Codex, verify their action contracts and download the source. Optionally publish and release them on GitHub.
• Spawner agent: install compatible GitHub packages as native processes or Docker containers. Supports pinned versions, stable releases, private repositories and private GHCR images.
• Agent catalog: discover available agents through MCP, or install directly from a GitHub repository URL.
• Mission builder: select available actions, define capabilities before an agent exists, configure optional actions and inspect saved missions as JSON.
• Managed secrets: encrypted persistence and delivery to agents, with masked Playground and Admin forms that keep values outside the model conversation.
• Task affinity: the same-agent rule keeps related actions on one agent instance.
• Execution checkpoints: actions can persist facts before external side effects and inspect them when handling retries.

Playground and Admin

• Admin UI is embedded in release binaries and Docker AIO.
• Completed operations link directly to their Admin details.
• Improved API-key retries, wheel/trackpad scrolling, text selection and terminal copying.
• Shared progress reporting keeps API, MCP, Playground and Admin consistent.

Fixes and maintenance

• Improved Windows filesystem messaging and test reliability.
• Updated vulnerable frontend dependencies.
• Replaced web/browser-agent integration dependencies with the lightweight example agent.

Upgrading from v0.1.0

• Replace mission root objective with objectives: ["Your objective"]. Step-level objective remains unchanged.
• Replace action requirement field secret with sensitive. Managed secret requirements use kind: "secret".
• Update MCP clients to read task counts and pending work from progress.
• Default deployments now start builder and spawner. Install web and browser agents when needed; they are maintained in separate repositories.
• Custom Go integrations must support the new mission creation and execution checkpoint interfaces.

Get started · Full changelog

Commander spawner agent v0.2.0

Commander spawner agent v0.2.0

Install published Commander agents as native processes or Docker containers
through the spawn-agent action.

Highlights

• Install any compatible GitHub agent package; catalog membership is optional.
• Use an exact version, @stable or an unversioned GitHub URL.
• Download checksum-verified native executables or launch digest-pinned Docker images.
• Access private GitHub repositories and GHCR images using GITHUB_TOKEN, supplied through the environment or managed-secret forms.
• Preserve selected versions and launch intents through execution checkpoints.
• Return the installed agent's identity and observed action readiness.

Native and Docker AIO deployments launch child processes. Compose launches sibling
containers on the deployment network.

Availability

Included in Commander's all-in-one and Compose deployments. Standalone packages
provide native binaries for Linux, macOS and Windows on amd64/arm64, plus a Linux
amd64 Docker image.

Package: github.com/defos-ai/commander/apps/agents/spawner@v0.2.0

Usage and configuration

Commander builder agent v0.2.0

Commander builder agent v0.2.0

Build Commander agents from a request, review their source and optionally publish
them for others to install.

Actions

• checkout-template: create a workspace from commander-example-agent.
• build-agent: generate code with Codex, run Go and package checks, verify registered action contracts and export a source.tar.gz artifact.
• publish-agent: publish reviewed source to a new or empty public/private GitHub repository.
• release-agent: run the repository's release workflow and return a verified installable package reference.

Failed build checks preserve the workspace for revision. Publishing and releasing
are optional; you can finish with the source artifact alone.

Supply Codex and GitHub secrets through Playground or Admin forms.

Availability

Included in Commander's all-in-one and Compose deployments. Standalone packages
provide native binaries for Linux, macOS and Windows on amd64/arm64, plus a Linux
amd64 Docker image.

Package: github.com/defos-ai/commander/apps/agents/builder@v0.2.0

Usage and configuration

Commander v0.1.0

Commander v0.1.0

The first release of DEFOS Commander: a framework and runtime for orchestrating
missions across distributed agents.

Highlights

• Run missions through chat: Playground connects your chosen model to Commander through MCP.
• Coordinate distributed agents: the controller discovers capabilities, checks readiness and assigns tasks.
• Automate follow-up work: mission rules collect results and hand decisions to code, models or people.
• Review execution in Admin: inspect operations, execution graphs, questions, reports and downloadable artifacts.
• Try website reviews: included web and browser agents check links, SEO, technologies, accessibility and screenshots.
• Choose your deployment: native all-in-one, Docker AIO or Docker Compose, with filesystem, MongoDB, RabbitMQ and S3 providers.

Native Commander binaries are available for Linux, macOS and Windows on amd64 and arm64.

All-in-one sessions are temporary. Use Compose to retain results.

Get started