Agents inside your product
Give users a durable project workspace. Submit tasks from your service and show progress through the Run event stream.
Sandpi Platform
Your model connection. Native agents. One API.
Add coding-agent work to your product with headless Sessions and Runs. Sandpi runs the native harness and persistent Sandbox; you bring the model API key and compatible provider or gateway. Stream progress, collect results, and return to the same project across tasks.
One Environment, many task histories
Share the workspace. Pay for one Sandbox.
Sessions retain separate native conversation histories while sharing the Environment's files, tools, configuration, and Sandbox. New Sessions reuse that Sandbox without a separate runtime charge. Use separate Environments when tasks need independent working state.
Repository automation
Let a pull-request update or CI failure start work in the Environment that already holds the project and its tools.
Recurring project work
Schedule maintenance tasks and collect results through the API, retaining native history and files between Runs.
Workflow
Environment → Session → Run.
Create an Environment
Give your service an isolated Sandbox with a persistent workspace. Sessions in that Environment share its files, tools, and compute.
Connect your model
Open a native harness Session with your model API key and compatible endpoint. Each Session keeps its own conversation history across Runs.
Submit a Run
Send a task, stream progress, and collect its result and file changes. Runs in the same Environment execute serially, across all Sessions.
Model calls use the harness's native protocol. Choose a compatible API format; Sandpi does not translate model APIs. Read the connection guide →
Choose your workflow
Sandpi for you. Platform for your software.
Both use persistent Sandbox0 Environments. Each product has its own Environments, Sessions, and billing.
| Your workflow | Sandpi · sandpi.ai | Sandpi Platform |
|---|---|---|
| How you work | Control a native agent terminal, files, and browser yourself. | Your service creates Sessions, submits Runs, and streams events through an API. |
| Agent access | Sign in to your agent or supply its credentials. | Bring your model API key and an endpoint compatible with the native harness. |
| What you pay for | A hosted Environment plan. Your provider bills model usage separately. | Shared Sandbox runtime, billed by the second. Your model provider or gateway bills model usage directly. |
| Where it runs | A cloud Environment, or your own connected machine. | A persistent cloud Environment for headless work. |
Harnesses
Use the agent you need.
Codex
Bring an endpoint that supports the OpenAI Responses API.
Claude Code
Bring an endpoint that supports the Anthropic Messages API.
Pi
Choose its native Responses, Chat Completions, or Anthropic Messages adapter for your endpoint.
Kimi Code
Use its native OpenAI-compatible Chat Completions connection.
ZCode
Use its native OpenAI-compatible Chat Completions connection.
Data controls
Choose retention. Track deletion. Inspect audit facts.
Run record retention
Choose 7, 30 or 90 days per Environment; the default is 30. Terminal Run instructions, output, events and file manifests expire automatically. Workspace files and native history persist until Environment deletion.
Deletion receipts
Erase Run records or request Environment online-content cleanup through the console or API. Durable receipts show pending, retrying and completed cleanup. Billing metadata, independent forks, backups and provider copies have separate retention.
Audit visibility
Query and export 90-day Platform operation metadata with user/API key attribution. Query Sandbox0 signed runtime facts when the deployment's audit feature is enabled. Runtime audit retention follows that deployment.
Deletion receipts cover online content and do not verify physical or backup erasure. Read the scope and API documentation, or contact us for backup retention, provider processing terms and a data processing agreement.
Usage-based pricing
Pay for runtime. Bring your model access.
Sandbox runtime is billed by the second. The fixed 2 vCPU / 4 GiB rate is equivalent to $0.10 per hour. The hourly figure is a price reference: pay for actual seconds, shared across Sessions. Paused time is free. Your provider or gateway bills model usage directly. Sandpi does not resell model tokens. Credit top-ups have no fee.