sandflare 2.1.13

Sandflare Python SDK — Firecracker microVM sandboxes for AI agents

Sandflare Python SDK

The official Python SDK for creating and managing Sandflare sandboxes.

Install

pip install sandflare

For local development:

pip install -e sdk/python/

Quick start

from sandflare import Sandbox

sandbox = Sandbox.create("agent")
sandbox.run_code("value = 10")
result = sandbox.run_code("value += 5\nprint(value)")
print(result.stdout)

Environment variables

• SANDFLARE_API_KEY: preferred API key
• SANDFLARE_API_URL: optional API base URL override
• PANDAAGENT_API_KEY: legacy API key alias
• PANDAAGENT_BASE_URL: legacy base URL alias

Sandbox methods

exec / run_python / run_node

result = sb.exec("ls /home/agent")
result = sb.run_python("print('hello')")
result = sb.run_node("console.log('hello')")

stream / exec_stream

stream() reads all SSE output then yields StreamEvent objects (backward-compatible):

for event in sb.stream("python3 train.py"):
    if event.event in ("stdout", "stderr"):
        print(event.line, end="", flush=True)
    elif event.event == "done":
        print(f"\nexited {event.exit_code}")

exec_stream() is a true SSE generator that yields events as they arrive:

for event in sb.exec_stream("python3 train.py", timeout=120):
    if event.type in ("stdout", "stderr"):
        print(event.data, flush=True)
    elif event.type == "done":
        print(f"\nexited {event.exit_code}")

metrics()

m = sb.metrics()
print(f"CPU: {m.cpu_used_pct:.1f}%  RAM: {m.mem_used}/{m.mem_total} bytes")

kill_process(pid)

sb.kill_process(1234)

git_clone(repo, ...)

result = sb.git_clone(
    "https://github.com/org/repo",
    path="/home/agent/repo",
    branch="main",
    depth=1,
)
print(result.output)

File I/O

sb.write_file("/home/agent/data.csv", csv_text)
content = sb.read_file("/home/agent/output.txt")
sb.upload("local.png", "/home/agent/img.png")
raw_bytes = sb.download("/home/agent/img.png")
entries = sb.ls("/home/agent")

Template class

Build custom sandbox templates from Dockerfiles:

from sandflare import Template

# Submit a build (returns immediately)
job = Template.build(
    name="my-template",
    dockerfile="FROM ubuntu:22.04\nRUN apt-get update && apt-get install -y python3",
    description="Custom Python env",
)
print(job.id, job.status)  # e.g. "tmpl-abc123", "building"

# Poll until ready (up to 30 minutes)
job = Template.wait_for_build(job.id, timeout_seconds=600)
print(job.status)  # "ready"

# Use template when creating a sandbox
sb = Sandbox.create("agent", template_id=job.id)

# List all templates
jobs = Template.list()

# Check build status manually
job = Template.get_build_status("tmpl-abc123")

# Delete a template
Template.delete("tmpl-abc123")

New dataclasses

Class	Fields
ProcessInfo	pid, command, cpu_percent, memory_percent
SandboxMetrics	sandbox_id, status, cpu_count, cpu_used_pct, mem_total, mem_used, disk_total, disk_used, ts
GitCloneResult	path, repo, branch, output
TemplateBuildJob	id, name, status, description, error, team_id, created_at, updated_at, logs

Build locally

cd sdk/python
python3 -m pip wheel . --no-deps -w dist/

Publish to PyPI

cd sdk/python
rm -rf dist/
python3 -m pip wheel . --no-deps -w dist/
python3 -m twine upload dist/*

Use a PyPI API token via TWINE_USERNAME=__token__ and TWINE_PASSWORD=<pypi-token>.