Changelog

What’s New

Model Availability Controls Organization and Project administrators can exclude models by model ID, provider, or vendor. Organization rules flow into every Project, and safety checks prevent a policy from removing every active model. See Model governance.

Custom Agent Eval Rubrics Agent eval runs can now use a custom objective and ordered rubric. The resolved policy is stored with the run, while evals without a custom policy continue to use functional equivalence. See Evals.

Project-Wide Share Management A unified Shares view brings Task and agent shares into one searchable, filterable list. Share tags can be managed from the same Project-level workflow.

CLI Access Tokens and Environments The Rightbrain CLI can return a refreshed access token with its deployment and Project context. JSON environment listings make authenticated API automation independent of the CLI’s private credential-file format. See Rightbrain CLI.

Improvements

• Oversized agent tool responses are replaced with a bounded omission record before they enter model-visible history
• Failed agent runs retain more context and report accurate total charged credits
• MCP servers support safer deletion, reauthorization, and bulk tool registration
• OpenAPI error contracts and request-body documentation now align with live API behavior
• New model options include GPT 5.6 Sol, Terra, and Luna, plus Gemini 3.6 Flash
• Agent chat validates file-size limits before starting a run