Terno Agent

A single-agent coding CLI and Python SDK. The agent reads and edits files, runs shell commands, executes Python in a sandbox, tracks its own task list, spawns subagents for parallel work, recalls persistent memory, and pulls in extra tools from any Model Context Protocol (MCP) servers you configure.

This repository is a monorepo: the agent itself lives in sdk/, and the rest is a reference web app and benchmark harness built on top of it.

Repository layout

Path	What it is	Details
sdk/	The terno-agent package — the terno CLI and the terno/terno_agent Python SDK. This is the core of the project.	sdk/README.md
backend/	FastAPI server that wraps the SDK, streams turns over a WebSocket, and persists chat sessions + messages to SQLite.	—
frontend/	React + TypeScript + Vite demo UI (landing + chat) that talks to the backend over WebSocket.	frontend/README.md
benchmarks/	Runners for Terminal-Bench (1.x and 2.0 via Harbor) and SWE-bench.	benchmarks/README.md
jobs/, runs/	Output directories for benchmark runs.	—

Installation

With uv (recommended)

uv add terno-agent

With pip

pip install terno-agent

Quick start

1. The CLI / SDK

The agent is fully usable on its own — no web app required.

cd sdk
uv tool install --editable ".[all]"     # installs the `terno` CLI on PATH
cp .env.example .env                     # then set ANTHROPIC_API_KEY or OPENAI_API_KEY

terno ask "refactor utils.py into smaller modules"
terno chat                               # interactive REPL

Or use it as a library:

from terno import Agent

with Agent.from_env() as agent:
    print(agent.run("summarize what this repo does").answer)

See sdk/README.md for the full surface: tools, permission prompts, sandbox backends, MCP config, Agent Skills, memory, and the SDK API.

2. The demo web app

The backend wraps the SDK as a WebSocket chat server; the frontend is a React UI that connects to it. Run them in two terminals.

Backend (FastAPI + SQLite, depends on the local sdk/ via an editable install):

cd backend
cp .env.example .env                     # set your LLM API key
uv sync
uv run uvicorn main:app --reload         # serves http://127.0.0.1:8000

Frontend (Vite dev server on port 5173, which the backend's CORS allows):

cd frontend
npm install
npm run dev                              # serves http://localhost:5173

3. Benchmarks

cd sdk
uv sync --extra benchmarks

Then run Terminal-Bench or SWE-bench as documented in benchmarks/README.md.

License

Apache 2.0 — see LICENSE.