Skip to content

v1.2.0

Choose a tag to compare

@tevfik tevfik released this 11 Mar 19:49
· 259 commits to main since this release

馃殌 Key Features

OpenAI-Compatible RAG Proxy (POST /v1/chat/completions)

Gleann can now act as a drop-in RAG proxy for any tool that speaks the OpenAI API protocol (Cursor, VS Code, LangChain, Python SDK, etc.).

  • Zero-config Routing: Set the model name to gleann/my-docs to automatically retrieve context from the my-docs index before forwarding to the LLM.
  • Multi-Index: Use gleann/docs,code to search across multiple indexes simultaneously.
  • Pure LLM Mode: Use gleann/ to bypass RAG and chat directly with your configured LLM.
  • Full Streaming Support: Native SSE chunk streaming for fast, real-time responses.
  • Endpoint: GET /v1/models natively exposes all your local gleann indexes as selectable AI models.

Interactive Conversation TUI & History

The CLI chat experience has been completely rewritten to support persistent state.

  • Conversation History: Chat sessions are now automatically saved and summarized.
  • Interactive Picker: Use gleann chat --pick to open a beautiful TUI to select, resume, or start conversations.
  • Index-less Chatting: You can now start a terminal chat session purely for LLM interaction without needing to select an active RAG index.
  • Word-Wrap Streaming: Terminal output now gracefully word-wraps to fit your terminal window width during streaming.

Multi-Index Search & Backend Upgrades

  • Multi-Searcher Engine: Powerful new ability to fan-out queries across multiple heterogeneous vector indexes simultaneously and rank merged results by score.
  • Embedding Cache: Vastly speeds up re-indexing by skipping embedding generation for unchanged document chunks.
  • gleannignore: Native support for .gitignore-style syntax to exclude specific files/folders during index builds.
  • Custom System Roles: Added pkg/roles registry to easily apply tailored system prompts (e.g., --role code or --role explain) from the CLI.

馃洜 Server & Telemetry Enhancements

  • Prometheus Metrics: Added /metrics endpoint exposing vector search latency, index sizes, build durations, and HTTP stats for Grafana dashboards.
  • Webhook Subscriptions: Added /api/webhooks to register external services for notifications on events like build_complete and index_deleted (with HMAC-SHA256 signature validation).
  • Graph Re-Indexing: Fixed duplicate PK constraints during KuzuDB AST graph re-indexing and added auto-detection of Go modules.

馃悰 Fixes & Housekeeping

  • Fixed TUI index picker syntax errors preventing proper skipping behavior.
  • Fixed trailing comma serialization issues in OpenAPI schema generation.
  • Corrected missing glamour dependency in go.mod causing CI/CD pipeline build failures.
  • CLI commands have been split and refactored into modular cmd_*.go files for maintainability.

馃摝 Upgrading

Install the latest version using go install:

go install github.com/tevfik/gleann/cmd/gleann@latest

Full Changelog: v1.1.6...v1.2.0