Navin Python SDK: Run an AI Agent from Python

This guide shows when to use the Navin Python SDK instead of calling a model directly. The SDK runs the same agent runtime used by the CLI: model routing, tools, workspace access, session history, memory, streaming events, and runtime helpers.

What you will build

a Python script that creates a Navin
one agent run from code
an optional streamed run with tool visibility

When to use this

Use the Python SDK for notebooks, evals, product backends, local scripts, workflow runners, and integrations that need direct access to agent sessions, memory, hooks, runtime state, or structured run results.

Use the OpenAI-compatible API instead when another language or process should call navin over HTTP.

Install

python -m pip install navin-ai
navin webui   # configure provider & model in the platform (Settings → Providers)
navin agent -m "Hello!"

Minimal working example

import asyncio

from navin import Navin

async def main() -> None:
    async with Navin.from_config() as bot:
        result = await bot.run("List the top-level files in this workspace.")
    print(result.content)

asyncio.run(main())

Production notes

Reuse one Navin instance for related work.
Pass session_key when a user, job, or eval case needs persistent history.
Use bot.stream(...) when the caller needs live text, tool, or failure events.
Use hooks for audit logs or custom observability.

Security notes

The SDK uses the same config, workspace, tools, and secrets as the CLI.
Do not run untrusted prompts with broad file or shell access.
Keep separate config/workspace paths for separate products or tenants.

Troubleshooting

If SDK code fails, first run navin agent -m "Hello!" in the same environment.
Print bot.runtime.workspace and bot.runtime.model to confirm the expected config loaded.
Use explicit config_path and workspace when scripts run from services.