mitshe

Mitshe runs AI coding agents in isolated Docker workspaces. Snapshot your dev environment once, clone it per task — multiple agents work in parallel without port conflicts or shared state. The problem: Running Claude Code (or any AI coding agent) on real codebases is risky. Agents touch your filesystem, conflict with other tasks, mess up your git state. Manual workarounds — git worktrees, separate VMs, copying repos — are tedious and break easily. The solution: Each agent session in mitshe runs in its own isolated Docker container with terminal, browser, and git. You configure the dev environment once (install deps, clone repos, set up databases with test data, run dev servers), snapshot it, and clone that snapshot for every new task. Five agents on five different tickets simultaneously, each with a clean copy of your env. How it works: - Spin up sessions from snapshots in seconds - Each session has its own terminal, file browser, and git - Chat-first interface — describe what you need, agents handle the rest - Visual workflow builder for automations: Jira ticket → branch → AI code review → tests → PR → Slack notification - Skills as reusable slash commands (/e2e-testing, /security-audit, /refactor, custom ones) - Polling-based integrations work behind NAT — no public URL needed - BYOK with AES-256 encryption for credentials - 40+ MCP tools exposed to the orchestration chat layer Supported AI providers: Claude (subscription via Claude Code, or API), OpenAI, Gemini, Groq, OpenRouter. Integrations: Jira, GitHub, GitLab, Slack, Discord, Telegram, Linear, ClickUp. Tech: Self-hosted in a single Docker container with SQLite. No external services, no managed cloud, no telemetry. MIT licensed. Built for: developers running Claude Code across multiple repos, teams automating repetitive dev workflows (code review, testing, ticket-to-PR pipelines), and self-hosters who want AI agents on their own infrastructure without giving up control over their data or environment. Use cases: - Run multiple Claude Code agents in parallel on different tickets - Automate "Jira issue → tested PR" workflows end-to-end - Sandbox AI agents safely on production codebases - Standardize dev environments across team members with shared snapshots - Build custom workflows mixing AI agents and CI/CD steps