tama 玉

Why tama?

You write agents as .md files. tama run runs them. tama brew ships them.

No frameworks. No orchestration runtime to manage. No Python glue.

13 built-in patterns

From a single LLM call (oneshot) to multi-agent debates and plan-execute loops. Declare the pattern — tama implements it.

Full observability

Every agent step, every LLM call, every artifact traced to DuckDB. Diff runs, replay with identical input, debug interactively.

Skills as Markdown

Tools are .md files following the Anthropic Agent Skills spec. Human-readable, git-diffable, composable.

Lean production deploys

tama brew compiles your agents into a distroless Docker image (~8MB). One command from laptop to cloud.

At a glance

# Scaffold
tama init my-project
cd my-project
tama add react my-agent
tama add skill search-web

# Iterate
ANTHROPIC_API_KEY=sk-... tama run "research fusion energy trends"

# Ship
tama brew
docker push my-project:latest

# agents/my-agent/AGENT.md
---
name: my-agent
description: Research agent with web search.
version: "1.0.0"
pattern: react
call:
  model:
    role: thinker
  uses:
    - search-web
---

You are a research assistant. Use `search-web` to find information.
When done, call finish with key="done" and a comprehensive summary.