This change makes system prompt text and tool changes part of the transcript rather than silently rewriting its starting conditions. This lets Pi record when instructions changed or tools became available, restore that state after resuming or navigating branches, and preserve cached prompt prefixes where the upstream supports it.
#7638 hardcoded vLLM's thinking_token_budget. Map thinkingBudgets through
thinkingTokenBudgetField, keep the boolean as a vLLM alias, clamp so the
answer keeps 1024 tokens, and document it.
fixes#8274
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Mario Zechner <badlogicgames@gmail.com>
Add the constrained sampling API and provider plumbing for OpenAI and Anthropic tool calls, including grammar sampling variants and model capability metadata.
Move provider auth and OAuth flows onto pi-ai Models, compose models.json and extension overlays through ModelRuntime, and retain ModelRegistry as an extension compatibility facade.
Adds a Context Overflow Errors subsection to custom-provider.md showing how a custom provider extension can rewrite unrecognized overflow errors via a message_end handler so pi's native compact-and-retry path triggers. Includes guard rails for provider scoping, idempotency, and avoiding rate-limit phrases.
- Document stream.push/end pattern instead of yield
- Link to existing provider implementations on GitHub
- Add testing section with list of test files to run
- Include proper content block and tool call examples
- Add resetApiProviders() to clear and re-register built-in providers
- Add createAssistantMessageEventStream() factory for extensions
- Add streamSimple support in ProviderConfig for custom API implementations
- Call resetApiProviders() on /reload to clean up extension providers
- Add custom-provider.md documentation
- Add custom-provider.ts example with full Anthropic implementation
- Update extensions.md with streamSimple config option
