Skip to content
diegokoes edited this page Jul 5, 2026 · 3 revisions

tachý wiki

Internal design and data-model reference for tachý. The README covers features, installation, deployment, and usage; this wiki covers how and why it is built the way it is.

tachý is a self-hosted, source-agnostic knowledge engine for engineering work items. It keeps an archive of approved lessons from past tickets and issues and retrieves them when a similar work item comes in. It never calls an LLM on its core path; the agent (Claude Code, Codex CLI, or any MCP client) is the reasoning layer. The one Claude-coupled component is the optional server-side agent behind the web Chat.

Two loops around work items, plus a third way in that skips the ticket entirely:

  • Ingest. Analyze a work item, then save an approved lesson. Nothing becomes knowledge until a human approves it.
  • Consult. Given a new work item, surface relevant past lessons.
  • Load context. Feed in freeform material (a codebase overview, a runbook, an architecture note) with no work item at all, stored as reference docs and searched alongside the lessons.

Pages

Frequently asked

Is the SDK only for Claude Code? No. Two integration surfaces exist, and only one is Claude-specific. The MCP server is vendor-neutral: any MCP client (Claude Code, Codex CLI, VS Code Copilot, or an organization's own model) drives it. The server-side agent behind the web Chat uses the Claude Agent SDK, a headless programmatic library (the same engine that powers Claude Code, but not Claude Code itself). That is the only Claude coupling in the project, isolated to one file.

Clone this wiki locally