kazam 1.3.0

Agent workspace and static site generator. Codebase indexing, task tracking, visual board — one Rust binary.
kazam-1.3.0 is not a library.

kazam

The tool your coding agent didn't know it needed.

Codebase indexing, task tracking, and a visual board — one Rust binary, no dependencies.

Why

Your agent re-reads files it already saw. It scans entire directories to find one function. It loses track of what it's done across sessions. And you watch all of this happen in a terminal you can barely follow.

kazam fixes the three things that make agent-assisted coding slower and more expensive than it should be:

Without kazam With kazam
Navigation Agent explores with find, grep, ls — burning tokens on every turn Two-tier anatomy index. Agent reads a 68-line summary, drills into the directory it needs
Tracking Work disappears into terminal scroll. No record of what was done or what's left Structured task tracking — add, claim, close, block. Persists across sessions
Visibility You read a terminal, or you don't Visual board with live-updating task status, anatomy, and activity log

Benchmarks

Tested on real codebases with real bug-fix tasks, comparing kazam-equipped vs vanilla Claude Code:

Repo Files Task Cost Speed
Open-source Node app 1,000 7 blind bug fixes 37% cheaper 19% faster
Internal monorepo 5,800+ UI click handler fix 17% cheaper 36% faster
Internal service repo ~900 Named config change tie tie

kazam's advantage scales with repo size and task ambiguity. Targeted tasks with named files are a wash. Navigation-heavy tasks in large codebases are where the savings compound.

Install

# Homebrew (macOS / Linux)
brew install tdiderich/tap/kazam

# Cargo (any platform with Rust)
cargo install kazam

# Bleeding edge
cargo install --git https://github.com/tdiderich/kazam

Quickstart — agent workspace

cd your-repo
kazam workspace init --agent claude

That's it. kazam scans your codebase, writes a two-tier anatomy index to .kazam/, installs Claude Code hooks, and writes workspace rules. Your agent now:

  1. Reads the anatomy index first instead of exploring with find/grep
  2. Tracks work with kazam track — tasks persist across sessions
  3. Logs activity so you can see what changed

Open the visual board:

kazam board

Live-updating task status, file anatomy, and activity log — served locally with auto-refresh on any .kazam/ change.

Quickstart — static sites

kazam init my-site && cd my-site
kazam dev . --port 3000    # → http://localhost:3000, live reload

Edit index.yaml. Save. The browser reloads. That's the loop.

kazam also builds beautiful static sites from simple YAML — 30+ themed components, three shell types (standard pages, print-ready documents, full-viewport decks), and zero runtime JS in the output. Let your agent write the content:

kazam wish deck --yolo "Q3 pipeline review"

One command, one populated deck. Works with Claude Code, Gemini CLI, Codex, and OpenCode.

Docs + live examples · Components · Themes · Deploy recipes

How the workspace works

Anatomy — persistent codebase context

kazam ctx scan walks your repo and builds a two-tier index:

  • Summary (.kazam/ctx/anatomy.yaml) — root files + top-level directory rollups with file counts, token estimates, and descriptions. Typically under 70 lines even for repos with thousands of files.
  • Detail (.kazam/ctx/anatomy/<dir>.yaml) — individual files in each directory, with per-file descriptions and token counts.

Agents read the summary first, then drill into the directory they need. No find. No grep. No wasted turns.

Task tracking — structured, persistent, session-spanning

kazam track add "Fix the auth middleware" --priority 1
kazam track claim kz-a1b2 --name claude
kazam track close kz-a1b2 --reason "patched token validation"
kazam track ready --json    # what's unblocked, sorted by priority

Tasks live in .kazam/track/tasks.yaml. They survive session restarts, context compaction, and agent handoffs. The workspace rules tell agents to close tasks as they go — not batch at the end.

Board — visual workspace

kazam board

A themed, auto-refreshing local dashboard showing task status, codebase anatomy, and activity. Built with kazam's own rendering engine. More natural than watching a terminal scroll.

Hooks — invisible wiring

kazam workspace init --agent claude installs three Claude Code hooks:

  • Session start — surfaces anatomy drift and ready tasks
  • Post-write — logs file modifications to the activity feed
  • Session stop — rescans anatomy and suggests enrichment

No workflow changes. The hooks fire silently and only surface output when something is actionable.

Security

~10 direct Rust crates, Cargo.lock committed, cargo-audit in CI, protected main, signed release tags. Full scope: SECURITY.md. Report vulnerabilities privately via the GitHub advisory form.

Contributing

PRs welcome — agent-assisted contributions explicitly encouraged. See CONTRIBUTING.md.

License

MIT — see LICENSE.