Agentweaver

Appearance

Project generation model settings ​

Project generation model settings let you pin the model used for project-shaping work without changing the default model used by normal agent runs. For the API see the reference; for the implementation flow see the deep dive.

When to use it ​

Use these settings when a project needs a different model for planning surfaces:

  • Blueprint generation model — the model that turns a written description into a team blueprint.
  • Workflow generation model — the model used when blueprint generation says no library workflow fits and Agentweaver drafts a custom workflow.
  • Outcome spec generation model — the model that drafts the coordinator's initial outcome plan before dispatch.

Leave a field blank to inherit the global generation default shown in the UI (apps/web/src/pages/ProjectSettingsPage.tsx:991). Blank values are saved as null (ProjectSettingsPage.tsx:646-648).

Model provider prerequisite ​

A model provider must be ready before Agentweaver starts AI work. This setup is separate from model selection.

The project status identifies one effective source:

  • GitHub Copilot with project scope
  • GitHub Copilot with platform scope
  • A custom-key provider with platform scope

If no source is ready, select Set up model provider. If setup is unavailable to you, contact a Platform Admin.

Step by step ​

  1. Open a project and go to Settings.
  2. In General → Generation models, fill one or more model ID fields: Blueprint generation model, Workflow generation model, or Outcome spec generation model.
  3. Click Save generation models. The browser sends PUT /api/projects/{id}/provider-settings with the generation fields and preserves the default provider settings (ProjectSettingsPage.tsx:636-650).
  4. If the save succeeds, the page shows Generation model settings saved.
  5. To inherit again, click Reset to inherit defaults. The UI saves the three generation fields as null (ProjectSettingsPage.tsx:991-1050).

What changes after save ​

  • The next Generate blueprint request for this project uses blueprint_generation_model after resolving it through GenerationModelOptions.ResolveBlueprintModel (apps/Agentweaver.Api/Endpoints/BlueprintEndpoints.cs:58).
  • If generated blueprint parsing signals that no library workflow fits, the workflow fallback uses workflow_generation_model in WorkflowGenerationRequest.GenerationModel (apps/Agentweaver.Api/Blueprints/BlueprintService.cs:554).
  • The next coordinator run in this project uses outcome_spec_generation_model when drafting the outcome spec (apps/Agentweaver.Api/Coordinator/CoordinatorRunService.cs:247).

Existing runs do not change; these settings affect future generation calls.

Error behavior ​

If a model id is not allowed, the settings save returns model_id is not allowed (apps/Agentweaver.Api/Endpoints/ProjectEndpoints.cs:274). If a configured generation model cannot be used later, the generation endpoint now surfaces provider-specific remediation instead of hiding the failure as generic validation: auth failures, rate limits, unavailable model lists, unavailable models, runtime misconfiguration, and transient provider failures are classified by AgentProviderException (packages/Agentweaver.AgentRuntime/Providers/AgentProviderException.cs:41).