- Added OpenCode Zen provider with API key-based authentication supporting multiple AI models.
- Added four new free models via OpenCode provider: glm-4.7-free, kimi-k2.5-free, minimax-m2.1-free, and trinity-large-preview-free.
- Added glm-4.7-flash model via Zai provider.
- Updated pricing and configuration for multiple models across Kimi, MiniMax, OpenRouter, Anthropic, Vercel AI Gateway, and Zai providers.
- Removed google/gemini-2.0-flash-exp:free from OpenRouter and stealth models from Vercel AI Gateway.
- Added Kimi Code provider with support for both OpenAI and Anthropic API formats.
- Added kimiApiFormat configuration option to allow users to select between 'openai' and 'anthropic' API formats for Kimi Code provider.
- Added Kimi API format setting in the tools configuration tab with descriptions for each format option.
- Refactored Anthropic provider authentication logic to support multiple API endpoints including Kimi's Anthropic-compatible API.
- Updated OpenAI completions provider to route direct Kimi models to dedicated Kimi provider implementation.
- Fixed rate limit issues with Kimi models by always sending max_tokens parameter, which Kimi uses for TPM calculation regardless of actual output.
- Added detection for Kimi models (kimi-code provider or moonshotai/kimi ID) to apply the max_tokens workaround.
packages/ai:
- fix: handle "sensitive" stop reason from Anthropic API
- fix: normalize tool call IDs with special characters for Responses API
- fix: add overflow detection for Bedrock, MiniMax, Kimi providers
- fix: 429 status is rate limiting, not context overflow
packages/tui:
- fix: refactored autocomplete state tracking
- fix: file autocomplete should not trigger on empty text
- fix: configurable autocomplete max visible items
- fix: improved table column width calculation with word-aware wrapping
packages/coding-agent:
- fix: preserve external config.yml edits on save (#1046 by @nicobailonMD)
- fix: resolve macOS NFD and curly quote variants in file paths
- Extracted delta event batching and throttling logic into AssistantMessageEventStream base class to eliminate duplication across test mocks.
- Changed access modifiers from private to protected in EventStream class to allow subclass customization of queue, waiting, and completion handling.
- Simplified MockAssistantStream implementations across five test files by removing custom event handling logic and delegating to AssistantMessageEventStream.
- Added protected methods deliver() and endWaiting() to EventStream to support subclass event delivery patterns.
- Implemented delta event type guard and merging utilities to consolidate event batching logic in one place.
- Added profile endpoint integration to resolve user email addresses when not available in usage payload.
- Implemented profile caching with 24-hour TTL to reduce redundant API calls for email resolution.
- Added fetchProfile function to retrieve Claude account profile information from the API.
- Added buildProfileCacheKey function to generate cache keys based on provider, token fingerprint, and base URL.
- Added resolveEmail function to handle email resolution with fallback to profile API and caching.
- Added automatic token refresh for expired Kimi OAuth credentials in usage provider.
- Imported refreshKimiToken function to enable token refresh capability.
- Enhanced error handling with validation for refresh token availability and try-catch block for refresh failures.
- Converted wrapper class properties to dynamic getters that delegate to underlying tool objects.
- Added optional chaining operators for safer null/undefined access in JSON schema property handling.
- Changed output metadata wrapper to use Object.create() with property descriptors instead of object spread syntax.
- Updated @bufbuild/protoc-gen-es from ^2.10.2 to ^2.11.0.
- Updated @types/bun from ^1.2.18 to ^1.3.7.
- Updated prettier from ^3.8.0 to ^3.8.1.
- Updated @sinclair/typebox from ^0.34.46 to ^0.34.48 across multiple packages.
- Updated @bufbuild/protobuf from ^2.10.2 to ^2.11.0 in packages/ai.
- Updated multiple dependencies including chalk, diff, file-type, zod, winston, recharts, and mime-types to their latest patch and minor versions.
- Added Kimi Code provider integration with OAuth device authorization flow and token management.
- Added four new Kimi Code models (kimi-for-coding, kimi-k2, kimi-k2-turbo-preview, kimi-k2.5) with reasoning support and 262K context window.
- Added kimiUsageProvider for fetching and caching Kimi Code API usage quota information.
- Added kimi-code login command to CLI for OAuth authentication with Kimi Code.
- Updated openai-completions provider to support Kimi-specific headers and cached token formats.
- Updated MiniMax-M2 model pricing: input 1.2->0.6, output 1.2->3, cacheRead 0.6->0.1.
- Extracted tool choice mapping logic into dedicated utility module for provider-specific format conversions.
- Refactored task summary generation to use template-based rendering instead of string concatenation.
- Migrated system prompt generation to use Handlebars templates with structured data instead of hardcoded strings.
- Extracted JTD to TypeScript conversion logic into dedicated utility module for schema documentation.
- Consolidated prompt template rendering across multiple modules using unified renderPromptTemplate function.
- Simplified type assertions in stream module to use generic OptionsForApi type instead of provider-specific types.
- Added session compaction auto-continue feature that automatically sends a synthetic continuation prompt after compaction completes when enabled in settings.
- Added tool output pruning system that intelligently removes verbose tool outputs from session history during compaction to reduce token usage while preserving critical information.
- Added short summary generation for compaction results, providing concise PR-style summaries of conversation changes alongside full summaries.
- Added session.compacting extension hook that allows extensions to customize compaction prompts, add context, and preserve metadata during summarization.
- Added synthetic message support to mark system-injected messages (like auto-continue prompts) and prevent them from being persisted to editor history.
- Added ToolChoice type and toolChoice parameter support across all AI providers (OpenAI, Azure OpenAI, Anthropic, Google) enabling fine-grained control over tool/function selection during LLM calls.
- Added toolChoice override capability to Agent.prompt() method and session prompt options allowing callers to control tool selection behavior per request.
- Added provider-specific tool choice mapping functions (mapAnthropicToolChoice, mapGoogleToolChoice, mapOpenAiToolChoice) to normalize tool choice formats across different LLM APIs.
- Removed kernel heartbeat/ping mechanism from PythonKernel, simplifying health monitoring by relying on direct isAlive() checks instead of periodic HTTP requests.
- Extracted type guard functions for improved type safety and code reusability across RPC and migration modules.
- Centralized AgentEvent type validation into a dedicated Set constant and type guard function in rpc-client.
- Replaced inline type assertions with type guard functions in RPC response and event handling logic.
- Refactored test utilities to use parseSessionEntries helper and type guard functions for cleaner test code.
- Added reasoning content configuration properties to OpenAI compatibility settings.
- Migrated historical tool call context to an XML-like structure.
- Clarified the message to explicitly warn models against mimicking tool calls.
- Enhanced context interpretation for Gemini 3 models.
- Replaced external JSON/JSONL parsing libraries with Bun's built-in JSON5 and JSONL APIs across all packages.
- Removed dependencies: json5, ndjson, get-east-asian-width, json-stringify-safe, split2, through2, and @types/ndjson.
- Replaced custom text width and ANSI wrapping implementations with Bun.stringWidth() and Bun.wrapAnsi() APIs.
- Updated Bun type definitions from ^1.2.17 to ^1.2.18 and bun-types from ^1.3.5 to ^1.3.7.
- Refactored JSONL parsing throughout codebase to use Bun.JSONL.parse() and Bun.JSONL.parseChunk() with improved buffer management and error handling.
- Updated test assertions to use Bun.stringWidth() for dynamic width calculations instead of hardcoded values.
- Fixed filtering of empty user text blocks for OpenAI-compatible completions.
- Fixed normalization of Kimi reasoning_content for OpenRouter tool-call messages.
- Added compatibility flags for reasoning content field handling and tool-call requirements.
- Updated model registry with new models (kimi-k2.5, solar-pro-3, qwen3-max-thinking) and adjusted pricing/token limits.
can1357