Skip to content

python sdk for coding agents

Build with your favorite coding agent

Run real coding agents against real repositories in real sandboxes.

Use Claude Code, Codex, OpenCode, or your own agent whenever you need it. Run the same agent in Docker for CI, Docker Sandboxes for local isolation, and future cloud backends for production.

sbx-agents is the missing SDK layer between autonomous coding agents and isolated execution environments.

Codex OpenCode Claude Code ready Docker SBX Plain Docker SDK-first

Not another LLM app framework

OpenAI Agents SDK, Pydantic AI, LangGraph, and similar frameworks help orchestrate LLM calls, tool schemas, and chat workflows.

sbx-agents orchestrates coding agents that operate on repositories: edit files, run tests, use MCP, produce diffs, and return artifacts from isolated sandboxes.

Need LLM app frameworks sbx-agents
Chat/tool graph first-class not the focus
Run Codex/OpenCode/Claude Code-style CLIs awkward first-class
Isolated repo workspace external concern core abstraction
Collect git diff hand-roll built in
Swap sandbox runtime hand-roll backend interface
Multi-agent coding workflows hand-roll core pattern

Get started

Install the package:

$ uv add sbx-agents

$ pip install sbx-agents

Run Codex or OpenCode in Docker SBX:

from pathlib import Path

from sbx_agents import Agent, Runner
from sbx_agents.backends import Codex
from sbx_agents.sandboxes import DockerSbxSandbox

agent = Agent(name="fix-tests", backend=Codex(model="gpt-5-codex"))

result = Runner.run_sync(
    agent,
    sandbox=DockerSbxSandbox(workspace=Path("."), branch="agent/fix-tests"),
    prompt="Fix the failing pytest suite.",
)

from pathlib import Path

from sbx_agents import Agent, Runner
from sbx_agents.backends import OpenCode
from sbx_agents.sandboxes import DockerSbxSandbox

agent = Agent(name="reviewer", backend=OpenCode(model="anthropic/claude-sonnet-4-5"))

result = Runner.run_sync(
    agent,
    sandbox=DockerSbxSandbox(workspace=Path("."), branch="agent/review"),
    prompt="Review this repository and produce a risk-ranked report.",
)

See Get started for the full walkthrough.

Or define a reusable runner in Python and launch it from the terminal:

$ sbxa run main:runner "Fix failing tests."

What you can build

CI fixer agents

Fix failing tests and open PRs from isolated branches.

Review fleets

Compare Codex, OpenCode, and custom agents against same repo.

Agent products

Launch one sandboxed agent per issue, incident, or customer request.

Tenant isolation

Build FastAPI services with workspace-per-tenant execution.

Multi-agent workflows

Reviewer, fixer, and verifier cooperate in one sandbox.

Multi-sandbox workflows

Move selected artifacts between isolated environments.

Supported backends

Type Backends
Agents Shell, Codex, OpenCode
Sandboxes DockerSandbox, DockerSbxSandbox

Learn more