Now: inspect the working foundation
The current release connects models, agent tasks, research, tools, and saved outputs. What runs in a particular installation depends on its models, add-ons, permissions, and configuration. Use the manual for the workflow and release notes for version-specific changes.
Model and tool paths
Connect local or compatible external models and choose which configured tools a task may use.
Visible work
Follow agent tasks and research progress, review approvals, sources, and saved artifacts.
Open source
Inspect the MIT-licensed code.
Next: make longer work easier to trust
These are areas for continued design and testing, not features already available in every installation. The order may change as contributors bring failures, examples, and proposals to the repository.
- 01
Recovery and review
Test restart, retry, approval, and side-effect boundaries in real agent work; make failures and recovery clearer to operators.
- 02
Execution Graph
Explore plans with explicit dependencies, outputs, permissions, and completion checks instead of relying only on a linear list of steps.
- 03
Knowledge and reusable workflows
Continue improving source-backed knowledge spaces and explore user-reviewed workflows that can be reused safely.
Longer term: control across an organization
A wider Agent Runtime and organizational Control Plane remain a direction, not a description of today's default install or a promised commercial edition. The open-source workspace is the foundation; policies, budgets, audit, and deployment controls need their own evidence and design.
Organization boundaries
Define how projects, roles, data access, and external connections should be governed across teams.
Operational evidence
Make costs, approvals, execution history, and failures inspectable without claiming controls that have not been verified.
Help decide what moves forward
An issue with a reproducible case or a proposed workflow is more useful than a feature name alone. Describe the task, the environment, what happened, and what a successful result would look like. We review scope in public before large changes.
OpenMake