Mediforce v1.0.0
Mediforce v1.0.0 marks the first stable release of our open-source platform for human-in-the-loop AI workflows in regulated clinical processes.
This release introduces:
- Per-step AI control modes — choose between No agent, Assist, Cowork, Human review, and Autonomous agent depending on the level of oversight each workflow step requires.
- AI-assisted workflow building — describe a process in natural language and let the workflow assistant create or modify steps and transitions, with changes reviewed before saving.
- Per-step model configuration and cost tracking — bring your own API key, select the appropriate model for each step, and track costs at the workflow-run level.
- Comprehensive audit trails — automatically capture inputs, outputs, models, workflow versions, user roles, approvals, and other execution evidence.
- Versioned, locked workflows — activated workflow versions are immutable, keeping every run traceable to the exact definition that produced it. Dry runs allow approval gates and workflow logic to be tested before using study data.
- Roles, scoped secrets, and isolated execution — control workflow access, restrict credentials to specific workflows, and run agents in isolated containers.
- Flexible automation and integrations — trigger workflows manually, on schedules, or through webhooks; connect MCP tools with per-step permissions; manage human review through task queues; and run workflows from the CLI or CI.
v1.0.0 provides a stable platform surface for building regulated AI workflows while leaving validation decisions and processes with each organization.
Mediforce is available under the Apache 2.0 license.