quelch 0.11.5

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, two service components, one declarative YAML:

• Quelch MCP (Q-MCP) — the MCP server agents talk to. Typically runs in Azure (Container Apps), but doesn't have to.
• Quelch Ingest (Q-Ingest) — the worker that pulls from each data source into Cosmos DB. Typically runs close to the data source — that's often on-prem when Confluence / Jira Data Center isn't reachable from Azure.

You usually run one Q-MCP and one or more Q-Ingest workers. Plus the quelch operator CLI for planning and deploying.

New here? Start with docs/getting-started.md — a step-by-step happy-path walkthrough from brew install to a deployed MCP server an agent can talk to.

Architecture overview

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

Q-Ingest can sit on-prem next to Confluence / Jira Data Center while Q-MCP runs in Azure — both write/read the same Cosmos account.

Q-MCP 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-friendly Q-Ingest — quelch generate-deployment writes docker-compose, systemd, or Kubernetes manifests so Q-Ingest can run next to your data sources; 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 state per Q-Ingest 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