Argustack

Ask AI about your Jira, Git, and GitHub — powered by local data, not cloud APIs.

Argustack downloads your project data into a local hub (Postgres + Neo4j + Qdrant), indexes your codebases with tree-sitter, and gives Claude direct access via 46 MCP tools. All data stays on your machine.

Was ticket PROJ-123 implemented as described? Who reviewed the PR and what was the feedback? How long will it take Sarah to fix this bug?

Features

• Jira / Jira Proxy / CSV — issues, comments, changelogs, worklogs, links, all custom fields
• Git — commits, per-file diffs, automatic issue cross-references
• GitHub — PRs, reviews, comments, releases, automatic issue cross-references
• External DB — schema introspection + read-only SQL against your app database
• Cross-source timeline — Jira + Git + GitHub events in chronological order
• Semantic search — find issues by meaning, not just keywords (pgvector)
• Task estimation — predict duration per developer based on actual history
• Knowledge graph — entity-relationship graph (issues, developers, modules, PRs) with impact analysis
• Code intelligence — Cursor-style RAG: tree-sitter AST + Neo4j call graph + Qdrant semantic search. Default embeddings via Ollama (nomic-embed-text, 768d); LM Studio and Voyage AI are opt-ins.
• Multi-tenant hub — one Postgres + Neo4j + Qdrant stack in ~/.argustack/, many workspaces inside it (CASCADE-isolated by workspace_id).
• 46 MCP tools — Claude queries your data via SQL + Cypher + vector search. Every tool advertises title + readOnly/destructive annotations so Claude Desktop can warn before write operations.
• IDE Plugin — kanban board for JetBrains IDEs where columns are Claude Code skills.
• 100% local — no cloud, no accounts, no telemetry.

Quick Start

npm i -g argustack

# 1. Bootstrap the hub (Postgres + Neo4j + Qdrant + pgweb + optional Ollama).
argustack init

# 2. Create a workspace and bind your sources.
argustack workspace add my-project
argustack add jira   --workspace my-project --url https://team.atlassian.net --email you@co.com --token ATATT...
argustack add git    --workspace my-project --path /path/to/repo
argustack add github --workspace my-project --owner org --repo repo --token ghp_...
argustack add code   --workspace my-project --root /path/to/repo

# 3. Sync data and (optionally) index code.
argustack sync --workspace my-project
argustack code index --project my-project

# 4. Connect Claude.
argustack mcp install

That's it. Ask Claude about your project.

argustack init auto-installs missing dependencies (Docker via OrbStack, Ollama with nomic-embed-text), resolves port conflicts, and applies the hub schema. For the full decision flow and every branch (auto-install vs manual, port conflicts, re-init detection, LLM probe error classification), see AGENTS.md → Init Flow.

IDE Plugin

Kanban board inside JetBrains IDEs (IntelliJ, WebStorm, PyCharm, …) where columns are Claude Code skills. Drag a task card to a skill column and Claude executes it.

• Cards are Markdown files in Docs/Tasks/
• Columns auto-discovered from .claude/skills/
• Workflows group skills into pipelines (plan, implement, test, review)

Download from GitHub Releases and install via Settings → Plugins → Install from Disk.

The CLI and IDE plugin are independent — the plugin works without the CLI.

Security

Argustack is a CLI tool with no backend, no cloud, no accounts. Credentials live in ~/.argustack/config.env on your machine (chmod 600). Data lives in PostgreSQL + Neo4j + Qdrant on localhost. Nothing is uploaded anywhere unless you opt into Voyage AI / OpenAI embeddings.

Documentation

• AGENTS.md — full CLI command reference, MCP tool reference (46 tools), init flow diagram, architecture, file map, hub schema, debugging
• llms.txt — LLM-friendly project summary (read this if you're Claude)
• CLAUDE.md — architecture, tech stack, workspace concept, contributor conventions

License

MIT