{"full_name":"tmc/langchaingo","owner":"tmc","name":"langchaingo","shard":"1","url":"https://github.com/tmc/langchaingo","description":"LangChain for Go, the easiest way to write LLM-based programs in Go","homepage":"https://tmc.github.io/langchaingo/","topics":["ai","go","golang","langchain"],"tag_pages":["ai","go","golang","langchain"],"language":"Go","license":"MIT","avatar":"https://avatars.githubusercontent.com/u/3977?u=ba15c36cd7e0fd2d5fd1a4eb4df77420618a6d9e&v=4","created_at":"2023-02-18T20:04:54Z","pushed_at":"2026-11-11T02:54:58Z","metrics":{"stars":9779,"forks":1166,"watchers":77,"issues":457,"prs":1067,"releases":22,"issues_open":239,"issues_closed":218,"prs_open":195,"prs_closed":207,"prs_merged":665,"discussions":34},"growth":{"day":{"stars":0,"forks":0,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[9773,9774,9775,9775,9775,9779,9779],"forks":[1165,1165,1165,1165,1165,1166,1166],"watchers":[null,77,77,77,77,77,77],"issues_open":[null,239,239,239,239,239,239],"prs_open":[null,195,195,195,195,195,195]},"boards":{},"same_owner":[],"issues":[{"number":1071,"title":"Community development","comments":37,"created_at":"2024-11-22T00:43:34Z"},{"number":820,"title":"vectorstores: Support chromem-go","comments":14,"created_at":"2024-05-05T19:40:50Z"},{"number":1110,"title":"Add Support for DeepSeek LLM Integration","comments":10,"created_at":"2025-01-22T01:06:23Z"},{"number":369,"title":"Splitting langchaingo into sub-modules","comments":10,"created_at":"2023-11-23T16:50:51Z"},{"number":119,"title":"[feature request] possible to use golang openai client? ","comments":10,"created_at":"2023-06-08T01:31:59Z"},{"number":1225,"title":"Agents executor can't finished before max iterations","comments":9,"created_at":"2025-04-14T08:58:29Z"},{"number":1103,"title":"Proposing a new package called Toolbox which facilitates using a group of tools","comments":9,"created_at":"2025-01-15T16:49:59Z"},{"number":580,"title":"Conversation With Memory","comments":9,"created_at":"2024-01-29T14:11:09Z"}],"discussions":[{"number":1470,"title":"How to output the thought process when building an agent?","comments":0,"category":"General","created_at":"2026-02-01T03:09:10Z"},{"number":1459,"title":"DuckDB based VectorStore?","comments":0,"category":"Ideas","created_at":"2026-01-07T20:44:15Z"},{"number":1458,"title":"Is there a WithSystemPrompt option in any of the llm's?","comments":0,"category":"Q&A","created_at":"2026-01-07T15:05:49Z"},{"number":1410,"title":"Is there a way to list all available models?","comments":0,"category":"Q&A","created_at":"2025-09-29T21:35:56Z"},{"number":1354,"title":"Streaming for the agent. Help needed","comments":1,"category":"Q&A","created_at":"2025-08-09T23:05:58Z"},{"number":1282,"title":"RFC: Making langchaingo More Modular, Extensible, and Developer-Friendly","comments":4,"category":"General","created_at":"2025-05-31T07:53:55Z"},{"number":1256,"title":"How intgerate googleapis/go-genai with langchaingo","comments":0,"category":"Q&A","created_at":"2025-05-16T18:06:42Z"},{"number":1212,"title":"Use Deepseek with langchaingo","comments":1,"category":"Q&A","created_at":"2025-04-08T05:54:26Z"}],"latest_release":{"name":"v0.1.15","tag":"v0.1.15","published_at":"2026-10-05T05:35:01Z","body":"A patch release that replaces three retired default models, adds typed\nreasoning content and OpenAI web search, and fixes a number of provider bugs.\n\n## Action required\n\n**googleai, vertex: default embedding model is now `gemini-embedding-001`.**\n`embedding-001` now returns 404. The new model produces 3072-dimension\nvectors that are not comparable with the old ones, so indexes built with the\nold default must be re-embedded (and fixed-size vector columns resized).\n\n**googleai, vertex: default chat model is now `gemini-2.5-flash`.**\n`gemini-2.0-flash` is retired. Gemini 2.5 thinks by default and thinking\ncounts toward `WithMaxTokens`; a very small budget can produce no output,\nwhich now returns `ErrNoContentInResponse` instead of panicking.\nMulti-turn tool calling with Gemini 3 is not yet supported (it needs thought\nsignatures); that is planned for the next release.\n\n**anthropic: default model is now `claude-sonnet-4-6`.**\n`claude-3-5-sonnet-20240620` is retired. Clients that set a model are\nunaffected.\n\n**Unkeyed struct literals.** No exported API was removed, but\n`llms.ThinkingConfig`, `llms.ContentChoice`, `llms.CallOptions` and\n`googleai.Options` gained fields, so unkeyed literals of them no longer\ncompile. Use keyed fields. `ContentChoice` JSON now includes a `Parts` key.\n\n## Behavior changes\n\n- bedrock (Nova): text blocks are joined into one choice instead of one choice per block.\n- mistral: non-streaming tool calls stay on the choice that made them, not `Choices[0]`.\n- googleai: an error from `StreamingFunc` is now returned (with a nil response).\n- anthropic: reasoning-only streaming works; a stream error event ends the call;\n  `temperature`/`top_p` are omitted where the model or thinking mode rejects them.\n- llms: `ToolCall` JSON round trips keep the function name and arguments;\n  wrongly typed fields now return a wrapped `json.UnmarshalTypeError`.\n- llamafile: `WithN` and `WithMinLength` no longer set `n_ctx` and `min_p`; use `WithNCtx` and `WithMinP`.\n- chains: no longer clear a streaming function set by the caller.\n- cohere: invalid input returns an error instead of panicking.\n- googleai: `GOOGLE_API_KEY` is used only when no other auth option is given.\n\n## New\n\n- `llms.WithWebSearch` for OpenAI (#1454).\n- Typed reasoning parts: `ThinkingContent`, `RedactedThinkingContent` and `ContentChoice.Parts`.\n- `ThinkingConfig.Effort` and `Display` (`WithThinkingEffort`, `WithThinkingDisplay`),\n  mapped for Anthropic and OpenAI reasoning models.\n- `(*ContentResponse).AssistantMessage` for building the next turn of a tool loop.\n- anthropic: preserves thinking and multi-part messages across turns; supports claude-fable-5.\n- `testing/llmtest.TestModel`, a conformance check for `llms.Model` implementations.\n\n## Fixes\n\n- anthropic: keep every part of system, AI and tool messages (#1456, #1468, #1010);\n  accept `redacted_thinking` blocks; replace the retired default model (#1431).\n- openai: zero-argument tool schemas include `properties` (#1446);\n  `reasoning_effort` only for models that accept it; no `temperature` for search-preview models.\n- googleai: send the API key with `WithHTTPClient`; report reasoning support for Gemini 2.5.\n\n## New Contributors\n\n* @majiayu000 made their first contribution in https://github.com/tmc/langchaingo/pull/1454\n\n**Full Changelog**: https://github.com/tmc/langchaingo/compare/v0.1.14...v0.1.15\n","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}