{"full_name":"emcie-co/parlant","owner":"emcie-co","name":"parlant","shard":"1","url":"https://github.com/emcie-co/parlant","description":"Parlant is an open platform for building customer-facing chat agents with AI. Describe how the agent should behave, and your coding agent implements it for you.","homepage":"https://www.parlant.io","topics":["llm","customer-service","customer-success","python","ai-agent","customer-experience","lead-generation"],"tag_pages":["ai-agent","llm","python"],"language":"Python","license":"Apache-2.0","avatar":"https://avatars.githubusercontent.com/u/160175171?v=4","created_at":"2024-02-15T20:16:15Z","pushed_at":"2026-11-15T18:21:03Z","metrics":{"stars":18335,"forks":1555,"watchers":99,"issues":179,"prs":655,"releases":36,"issues_open":12,"issues_closed":167,"prs_open":2,"prs_closed":107,"prs_merged":546,"discussions":47},"growth":{"day":{"stars":3,"forks":-1,"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":[18329,18333,18333,18333,18333,18332,18335],"forks":[1555,1556,1556,1556,1556,1556,1555],"watchers":[null,99,99,99,99,99,99],"issues_open":[null,12,12,13,13,12,12],"prs_open":[null,2,1,1,1,2,2]},"boards":{},"same_owner":[],"issues":[{"number":683,"title":"[Docs] Warn users when NLP service does not offer constrained sampling","comments":7,"created_at":"2025-12-21T08:08:30Z"},{"number":819,"title":"[Bug] <SDK accepts cross-journey transitions and later crashes journey projection with `KeyError`>","comments":5,"created_at":"2026-06-23T16:01:06Z"},{"number":818,"title":"[Enhancement] Add Valkey as a VectorDatabase backend","comments":3,"created_at":"2026-06-22T15:52:55Z"},{"number":829,"title":"fix: embedding cache migration preserves the legacy document version","comments":1,"created_at":"2026-07-26T09:50:39Z"},{"number":236,"title":"[Enhancement] Guideline Statistics","comments":1,"created_at":"2025-01-16T15:27:06Z"},{"number":884,"title":"[Enhancement] Add a configurable relevance floor to ToolRecaller's remaining-tools bucket","comments":0,"created_at":"2026-11-12T15:59:13Z"},{"number":813,"title":"[Enhancement] Better templating","comments":0,"created_at":"2026-06-16T09:29:50Z"},{"number":691,"title":"[Enhancement] Activate Structured Outputs for AWS models","comments":0,"created_at":"2025-12-23T06:25:59Z"}],"discussions":[{"number":881,"title":"Delta frames for streaming events: what should the wire format be?","comments":1,"category":"Ideas","created_at":"2026-11-12T00:29:04Z"},{"number":838,"title":"Is Parlant 4.0 planned, and what's the plan for latency/performance optimization?","comments":0,"category":"Q&A","created_at":"2026-10-26T12:37:15Z"},{"number":804,"title":"Which model are you using for parlant ?","comments":0,"category":"General","created_at":"2026-05-20T01:08:03Z"},{"number":704,"title":"Integrate Parlant with Pipecat","comments":2,"category":"Q&A","created_at":"2025-12-31T11:51:13Z"},{"number":650,"title":"Using llama.cpp with parlant","comments":1,"category":"General","created_at":"2025-11-24T02:55:50Z"},{"number":621,"title":"Selecting the Best LLM or SLM Model for Parlant [Structured JSON]","comments":1,"category":"General","created_at":"2025-10-30T05:33:31Z"},{"number":616,"title":"Configuration: How to run examples with other LLM backend","comments":3,"category":"Q&A","created_at":"2025-10-27T12:11:20Z"},{"number":613,"title":"RFC: Multi-Agent Orchestration Layer for Parlant","comments":3,"category":"Ideas","created_at":"2025-10-26T12:58:07Z"}],"latest_release":{"name":"v4.0.5","tag":"v4.0.5","published_at":"2026-11-12T20:30:23Z","body":"## [4.0.5] - 2026-11-12\r\n\r\n### Added\r\n\r\n- Gemini 4 Hydrogen works through the Gemini provider, including streamed replies,\r\n  tool calls, saved conversation replay and token accounting. Select it as\r\n  `gemini:gemini-4-hydrogen`; existing default models are unchanged.\r\n- The chat widget shows Markdown tables in an agent's reply as tables. They\r\n  used to show as lines of pipes. Column alignment, inline Markdown in cells\r\n  and escaped pipes work. A streamed table appears once its header and divider\r\n  rows arrive and fills in cell by cell. A wide table scrolls sideways inside\r\n  the message, and your site's own table styles don't change it. Your project\r\n  keeps the copy `parlant widget init` made. To pick this up, run\r\n  `parlant widget init --force`, which replaces that copy and any changes you\r\n  made to it.\r\n- Event streams (`sse=true`) now carry an `id` on every frame and resume from\r\n  it. A client that reconnects with `Last-Event-ID` picks up after the frame it\r\n  last saw, instead of replaying the session from the beginning. Streams ask a\r\n  reconnecting browser to wait three seconds (`retry: 3000`), and an idle\r\n  stream sends a keep-alive comment every 15 seconds, so a proxy no longer\r\n  closes a quiet connection before `wait_for_data` expires. When a request\r\n  carries both `min_offset` and `Last-Event-ID`, the later of the two wins: a\r\n  reconnecting client usually still has the original `min_offset` in its URL,\r\n  and honoring it would replay the session. The delay and the keep-alive\r\n  interval can be set with `PARLANT_SSE_RETRY_MS` and\r\n  `PARLANT_SSE_HEARTBEAT_SECONDS`; `0` disables the keep-alive, and `0` for the\r\n  delay omits the `retry:` line.\r\n- Gemini serves Google's Gemini Decision 1 (`gemini-decision-1`), which answers\r\n  many yes/no, choice and scale questions about one conversation in a single\r\n  request. A `{\"*\": \"gemini\"}` route now uses it for rule ranking at the\r\n  efforts that rank rules in one batch, as OpenAI and OpenRouter routes already\r\n  do with their decision models. Name it on a route with\r\n  `\"matching.rank[capability=decision]\": \"gemini:gemini-decision-1\"`. It uses\r\n  your existing `GEMINI_API_KEY`. Gemini reports no cost for it, so add a\r\n  `gemini-decision-1` entry to `server.set_pricing(...)` (Google lists $0.10 per\r\n  million input tokens and nothing for output) to have it priced.\r\n\r\n### Changed\r\n\r\n- Gemini models count tokens on your server instead of asking the Gemini API,\r\n  which took about 300ms per count. A step after a tool call made two such\r\n  calls in a row while choosing canned responses, so replies after a tool call\r\n  now start about 0.4–0.6s sooner. Counts use the open Gemma 3 tokenizer and\r\n  are within a token or two of Google's. The `gemini` extra now installs\r\n  `sentencepiece`. The tokenizer downloads once from raw.githubusercontent.com,\r\n  in the background, on the first count. Until it's ready, or if your server\r\n  can't reach GitHub, Parlant counts with the API as before and logs one warning\r\n  saying so.\r\n- Agents no longer repeat what the customer just said back to them before\r\n  replying (\"Lead generation for a logistics startup, so…\"). A reaction now\r\n  responds to what the message means, and when the customer is answering the\r\n  agent's own question, the agent goes straight to its next step. The prompt no\r\n  longer bans stock acknowledgments such as \"makes sense\" by name.\r\n- Agents created by `parlant init` no longer learn to repeat the visitor's words\r\n  back. The warm-host and upbeat personalities now react to what a message\r\n  means, and the warm host goes straight to its next step when the visitor is\r\n  answering its question. Existing projects keep their own text; copy the new\r\n  wording from a fresh `parlant init` if your agent echoes.\r\n\r\n### Fixed\r\n\r\n- On managed (Dashboard-hosted) servers, `session.metadata.get()`, `.set()`\r\n  and `.delete()`, `session.update()` and `server.get_session()` failed with\r\n  \"Session not found\" for live conversations, b","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}