ZANA The coding agent that runs itself
Import your Claude Code and Codex history into local memory. Delegate to background agents that run while your terminal is closed.
bring your own keys — Anthropic · OpenAI · Ollama & more · no signup, no telemetry
Animated demo of Zana across eight use cases: steering the agent mid-run, fixing a legacy .NET service, analyzing microservices in parallel, debugging a Kubernetes crashloop, turning logs into a human report, a read-only autonomous agent, building your own typed tools, and organizing files beyond code. Use the buttons below to pick a case.
● latest v0.52.0 · changelog →
› under the hood
Terminal TUI
Streaming chat, live tool output — steer it mid-run, just keep typing.
~30 typed tools
Every tool declares its side effects. Machine contracts, not blind shell.
Persistent memory
Plain files on your disk. Imports your Claude Code and Codex history.
Background agents
Delegate, schedule, queue — the daemon runs them while the TUI is closed.
Toolpacks
Drop one file in ~/.zana-agent/toolpacks/ to add tools. No core changes.
Capability safety
Readonly mode + per-agent tool allowlists — you decide what each agent can touch.
› inside the TUI
- /model switch provider + model mid-session
- /memory inspect what it remembers, and why
- /cron schedule jobs from chat
- /agents autonomous agents — last / next run
- /compact free context, keep the plan
- /usage context, cost, cache
Named for the zana — the mountain fairy of Albanian folklore who guards those who work alone.
built in the open by Ordis Hysa ↗
independent project — not affiliated with Anthropic or any model provider