quelch 0.9.4

What is Quelch?

Quelch is a knowledge-platform operator tool for teams using Jira and Confluence. It ingests data into Cosmos DB as the system of record, uses Azure AI Search (via the embedded rigg library) for hybrid semantic retrieval, and exposes a five-tool MCP API that agents (Copilot Studio, VS Code Copilot, Claude, Codex) can call directly.

One Rust binary, one YAML config file, three runtime roles: quelch ingest, quelch mcp, and the operator CLI.

Architecture overview

                                              ┌─────────────────────────────┐
                                              │  Azure AI Search            │
                            ┌─────────────┐   │   ├─ Indexer (auto-vector)  │
Sources ──quelch ingest──►  │  Cosmos DB  │◄──┤   └─ Knowledge Base         │
                            │ (raw JSON)  │   │      (Agentic Retrieval)    │
                            └──────┬──────┘   └─────────────┬───────────────┘
                                   │ query · get             │ search
                                   │ aggregate               │ (semantic)
                                   ▼                         ▼
                                 ┌────────────────────────────┐
                                 │        quelch mcp          │
                                 │ (per-tool routing; 5 tools)│
                                 └─────────────┬──────────────┘
                                               │  MCP Streamable HTTP
                                               ▼
                              Agent (Copilot Studio / VS Code / Claude / Codex / …)

The MCP server fans out per tool: query/get/aggregate hit Cosmos DB directly (exact, exhaustive); search routes through the Azure AI Search Knowledge Base (Agentic Retrieval — question decomposition, reranking, optional answer synthesis); list_sources answers from a cached schema catalog without any backend call. See docs/architecture.md for full details.

Features

• Cosmos DB as system of record — exact queries, counts, exhaustive listings, and cursor-based pagination without hitting search
• Azure AI Search via rigg — indexes, skillsets, indexers, knowledge sources, and knowledge bases all managed from quelch.yaml
• Five-tool MCP API — search (Knowledge Base agentic retrieval), query (Cosmos SQL), get (point-read), list_sources, aggregate
• Incremental sync — minute-resolution windows with safety lag, backfill resume, soft-delete reconciliation
• Agent bundle generator — quelch agent generate produces grounded bundles for Copilot Studio, Claude Code, VS Code Copilot, Copilot CLI, Codex, and Markdown
• On-prem artefacts — quelch generate-deployment writes docker-compose, systemd, or Kubernetes manifests; Quelch never SSHes anywhere
• Operator CLI — azure plan, azure deploy, azure indexer, azure logs with Bicep + az shell-outs
• Rich TUI — fleet dashboard showing live ingest state per worker, polling quelch-meta

Installation

Homebrew (macOS/Linux)

brew install mklab-se/tap/quelch

Cargo

cargo install quelch

Binary download

Download pre-built binaries from the latest release.

Quick check

After installing, confirm the binary works and try Quelch entirely offline against the in-process simulator:

quelch --version
quelch dev          # offline — sim + Cosmos mock + ingest + MCP, all in one process

The TUI fleet dashboard appears; press q to exit. No Azure account or source credentials needed for quelch dev.

Getting started

When you're ready to run Quelch against real Jira / Confluence and deploy to Azure, follow docs/getting-started.md — a step-by-step happy-path walkthrough covering prerequisites, quelch init, planning, deploying, and connecting an agent.

CLI surface

Run quelch --help for the live command list. See docs/cli.md for every command and flag with examples and discussion.

Documentation

License

MIT