- Moved sticky positioning from mobile-only to global CSS for .top-left-actions
- Buttons now stay fixed at top above content on all screen sizes
- Includes various other app updates (version management, server improvements)
Documents the issue where tool calls count as separate Chutes AI requests,
proposed solutions, technical analysis, and user concerns about breaking
sequential workflows.
Includes:
- Root cause analysis of Vercel AI SDK multi-step execution
- 4 proposed solution options with pros/cons
- User concerns about model context and workflow breaks
- Code references and technical diagrams
- Recommended next steps for testing and implementation
Relates to: Tool call execution flow in session management
- Add 'cohere' to DEFAULT_PROVIDERS and PLANNING_PROVIDERS arrays in admin.js
- Add Cohere option to provider limits dropdown in admin.html
- Add 'cohere' to DEFAULT_PROVIDER_SEEDS, PLANNING_PROVIDERS, and KNOWN_USAGE_PROVIDERS in server.js
Cohere was already integrated in the opencode backend (SDK installed, provider registered, icons defined). These changes make it available in the admin panel for build message configuration.
- Parse model names like 'chutes/deepseek-ai/DeepSeek-V3.2-TEE'
- Extract provider prefix before '/' to detect chutes, cerebras, etc.
- This fixes provider detection when admin panel sets wrong provider
- Support providers as array of strings: [opencode, chutes]
- Support providers as array of objects: [{provider: opencode}]
- This fixes provider detection for Chutes and other providers
- Move loadAdminModelStore() before loadState() in bootstrap()
- This ensures adminModels is available when ensureOpencodeConfig is called
- Fixes issue where Chutes provider was not being configured
- Add check to load adminModels if not already loaded
- Add error handling for models.dev fetch failures
- Log provider source (models.dev vs built-in) for debugging
- Add fetchModelsDevProviderModels() function with 1-hour caching
- Update ensureOpencodeConfig() to dynamically load all models for configured providers
- Support Chutes, Cerebras, Groq, Google, and Nvidia providers
- Only fetch models for providers actually used in adminModels
- Cache models.dev data to reduce API calls and improve performance
This allows any model from models.dev to work when added in the admin panel,
instead of requiring hardcoded model definitions.
When models from providers like Chutes or Cerebras were added via the admin panel,
they didn't work in the builder page because the opencode.json config file only
included the Ollama provider configuration.
This fix adds Chutes and Cerebras provider configurations to the opencode.json
file when their respective API keys are available in environment variables
(CHUTES_API_KEY and CEREBRAS_API_KEY).
Fixes the issue where chutes/deepseek-ai/DeepSeek-V3.2-TEE and similar models
would fail to load from the builder page.
- Add multiple fallback global variable names for Editor.js plugins
- Add debug logging to track plugin availability
- Add error handling for editor initialization
- Ensure at least paragraph tool is available before initializing
- Fix ReferenceError: List is not defined when opening post modal
- Removed Docker build job from workflow (image built when deploying)
- Updated paths filter to only trigger on opencode/** changes
- Added artifact cleanup to keep only latest CLI build
- Added multi-stage Dockerfile to build CLI from source during Docker build
- Simplified permissions (removed packages write)
