Sandpi Platform

Coding agents, ready for your API.

Run a real coding-agent harness against a persistent project. Keep files and native agent history across Runs, stream progress, and return to the same Environment after it pauses.

Workflow

Environment → Session → Run.

01

Create an Environment

Give your agent a persistent Sandbox0 workspace with project files and a writable RootFS.

02

Choose a harness

Open a Session for Codex, Claude Code, or Pi. Each Session retains its native conversation history.

03

Submit a Run

Send one task, follow its event stream, and inspect the result and itemized model usage.

Harnesses

Use the agent you need.

01

Codex

Runs against the official OpenAI API.

02

Claude Code

Runs against the official Anthropic API.

03

Pi

Select a supported provider and model when you create a Session and submit a Run.

Usage-based pricing

Official model token rates.

Platform supplies the upstream API keys. Model tokens are billed at official provider prices. A 5% fee applies only when topping up prepaid credits, with no separate Sandbox charge.

Open Platform