# Tensorlake

> Tensorlake gives agents a stateful execution environment — isolated MicroVM sandboxes that boot in hundreds of milliseconds (with memory and filesystem preserved across suspend/resume) plus durable Orchestration for long-running, crash-resumable agent workflows. Works with any LLM provider (Anthropic, OpenAI), framework (LangChain, OpenAI Agents SDK, Claude Agent SDK), database, or API.

## Getting Started

- [START HERE: Tensorlake Skills for Coding Agents](https://docs.tensorlake.ai/agent-skills.md): Install the Tensorlake skill into Claude Code, Codex, or any coding agent so it learns to build production workflows with the Orchestration and Sandbox SDKs — the fastest way to get an agent productive with Tensorlake.
- [Authentication](https://docs.tensorlake.ai/platform/authentication.md): Learn how to make API requests to the Tensorlake APIs
- [Set up BYOC on AWS](https://docs.tensorlake.ai/platform/self-hosted-compute/aws.md): Run Tensorlake sandboxes in your AWS account, with executors on Amazon EC2 and either a Tensorlake-run or a self-hosted Sandbox Proxy.
- [Sandboxes Quickstart](https://docs.tensorlake.ai/sandboxes/quickstart.md): Install the SDK, authenticate, and run your first sandbox in under five minutes.
- [Orchestration + Sandboxes](https://docs.tensorlake.ai/applications/sandboxes.md): How Orchestration coordinates Sandboxes — the two core integration patterns: agent-in-sandbox and sandbox-as-tool.
- [Orchestration Quickstart](https://docs.tensorlake.ai/applications/quickstart.md): Write, deploy, and call your first Tensorlake Application, a serverless agentic web-scraper with the Claude Agent SDK in under five minutes.
- [SDK Reference (Orchestration)](https://docs.tensorlake.ai/applications/concepts.md): Functions, applications, decorators, request context, and lifecycle reference
- [Code Interpreter Agent](https://docs.tensorlake.ai/examples/agentic-applications/code-interpreter.md): Build a secure code execution environment using Tensorlake and OpenAI.

## Pricing

- [Pricing and billing](https://docs.tensorlake.ai/platform/billing.md): Compare Tensorlake plans, per-sandbox resource limits, concurrency, session duration, and metered compute and storage rates.

## Webhooks

- [Webhooks](https://docs.tensorlake.ai/platform/webhooks/overview.md): Receive signed, project-scoped notifications for sandbox, application, and application-request lifecycle events.
- [Configure Webhooks](https://docs.tensorlake.ai/platform/webhooks/configuration.md): Create and manage a project webhook destination, HTTPS endpoint, event subscriptions, and delivery status.
- [Webhook Event Reference](https://docs.tensorlake.ai/platform/webhooks/events.md): Event names, delivery envelope, payload fields, lifecycle semantics, and examples for Tensorlake webhooks.
- [Verify Webhook Signatures](https://docs.tensorlake.ai/platform/webhooks/signature-verification.md): Verify the Svix signature on every Tensorlake webhook before accepting or processing its payload.
- [Test a Webhook](https://docs.tensorlake.ai/platform/webhooks/testing.md): Send a signed synthetic lifecycle event from Tensorlake Cloud to validate a webhook destination.

## Sandboxes

### Core Docs

- [Sandbox and Orchestration Infrastructure for Agents](https://docs.tensorlake.ai/sandboxes/introduction.md)
- [SDK Reference (Sandboxes)](https://docs.tensorlake.ai/sandboxes/sdk-reference.md): Sandbox handle — every method, parameter, and data model you need to build with sandboxes, in one page.
- [Async SDK (Python)](https://docs.tensorlake.ai/sandboxes/async.md): Use the AsyncSandbox class to drive sandboxes from asyncio code.
- [Lifecycle](https://docs.tensorlake.ai/sandboxes/lifecycle.md): Sandbox states, creation, suspend/resume, and cleanup
- [Sandbox Pools](https://docs.tensorlake.ai/sandboxes/pools.md): Pre-warm sandboxes with pools for near-instant startup
- [Environment Variables](https://docs.tensorlake.ai/sandboxes/environment-variables.md): Set per-command and per-PTY environment variables in the CLI, Python, and TypeScript.
- [Commands & Processes](https://docs.tensorlake.ai/sandboxes/commands.md): Run commands, capture and stream output, and start, manage, and signal background processes
- [Sandbox Process Logs](https://docs.tensorlake.ai/sandboxes/process-logs.md): View, search, and filter stdout and stderr from sandbox processes in the Tensorlake Console
- [SSH and PTY Sessions](https://docs.tensorlake.ai/sandboxes/pty-sessions.md): Reach a running sandbox over standard SSH, or open a programmatic PTY session over WebSocket
- [File Operations](https://docs.tensorlake.ai/sandboxes/file-operations.md): Copy, read, write, and manage files inside Tensorlake sandboxes. Transfer between your local machine and the sandbox filesystem over the proxy URL.
- [Mount Filesystems](https://docs.tensorlake.ai/sandboxes/mount-filesystems.md): Mount durable, shareable Tensorlake filesystems into sandboxes at boot, on warm-pool claims, or on a running sandbox, optionally read-only, prefetched, pinned to a permanent snapshot, or owned by a guest user of your choice.
- [Snapshot and Clone](https://docs.tensorlake.ai/sandboxes/snapshots.md): Save, restore, and clone sandbox filesystem, memory, and running processes
- [Tensorlake Images](https://docs.tensorlake.ai/sandboxes/tensorlake-images.md): The managed tensorlake/* images, what ships in them, and how they behave when you run them.
- [Build and Import Images](https://docs.tensorlake.ai/sandboxes/images.md): Import or build custom images and run them in sandboxes.
- [Computer Use](https://docs.tensorlake.ai/sandboxes/computer-use.md): Launch ubuntu-vnc sandboxes and drive their desktop from Python or JavaScript.
- [Skills in Sandboxes](https://docs.tensorlake.ai/sandboxes/skills-in-sandboxes.md): Pre-load TensorLake skill files inside sandbox images so coding agents auto-discover them at startup.
- [Networking](https://docs.tensorlake.ai/sandboxes/networking.md): Route internet traffic into sandbox applications and control outbound internet access
- [Local Tunnels](https://docs.tensorlake.ai/sandboxes/tunnels.md): Forward a local TCP port to a port inside a sandbox over an authenticated WebSocket.
- [Run Docker](https://docs.tensorlake.ai/sandboxes/docker.md): Run Docker containers inside Tensorlake sandboxes using the tensorlake/ubuntu-systemd base image, with full systemd support for compose, networking, and daemons.
- [Cloud Volumes](https://docs.tensorlake.ai/filesystems/introduction.md): Cloud Volumes are directories that can be mounted on one or more sandboxes. Volumes can be versioned and can be time-travelled.
- [Core Concepts](https://docs.tensorlake.ai/filesystems/core-concepts.md): Short definitions for the versioned file system model: mounts, local journals, autosave checkpoints, permanent snapshots, retention, and publishing.
- [Git Repositories](https://docs.tensorlake.ai/git/introduction.md): Managed Git repositories built for agents that scales to tens of millions of repositories and hundreds of commits per second per repository.
- [Git Repositories](https://docs.tensorlake.ai/git/git-repositories.md): Scalable Tensorlake-hosted git repositories for agents
- [Repository SDKs](https://docs.tensorlake.ai/git/repository-sdks.md): Create repositories, inspect refs, push worktrees, and merge branches from Python and TypeScript.
- [File System Mounts](https://docs.tensorlake.ai/filesystems/filesystem-mounts.md): Choose the right versioned file system mount for agent work, fixed inputs, and shared assets.
- [Repository Mounts](https://docs.tensorlake.ai/git/workspace-mounts.md): Mount a repository lazily, autosave into private workspace WAL, materialize deliberate commits, and land changes with server-side promote and rebase.
- [Read-only Mounts](https://docs.tensorlake.ai/filesystems/read-only-mounts.md): Give sandboxes read-only views of shared file systems, and pin mounts to permanent snapshots for reproducible inputs.
- [Manage Sessions](https://docs.tensorlake.ai/filesystems/manage-sessions.md): Inspect, resume, recover, and clean up versioned file system sessions.
- [Concurrent Writes](https://docs.tensorlake.ai/filesystems/concurrent-writes.md): How several mounts writing to one file system reconcile. Disjoint paths merge automatically, same-path writes are last-writer-wins.
- [Merging Changes](https://docs.tensorlake.ai/git/merging.md): Land workspace changes on branches that moved, rebase a workspace, resolve conflicts, and merge branches directly.
- [Commits](https://docs.tensorlake.ai/git/commits.md): List commit history with landing attribution, filter it by file or directory, read per-commit diffs, and inspect workspace and repository graphs from the CLI or over HTTP.
- [Blame](https://docs.tensorlake.ai/git/blame.md): Attribute every line of a file to the commit that introduced it, with landing attribution on each range, over HTTP.
- [Authentication](https://docs.tensorlake.ai/git/authentication.md): Authenticate the Tensorlake CLI, plain Git clients, and repository mounts.
- [Architecture](https://docs.tensorlake.ai/filesystems/architecture.md): How versioned file systems and repositories store metadata, deliver file content, merge changes, and expose operational state.

### Use Cases

- [Store Agent-Generated Code](https://docs.tensorlake.ai/git/store-generated-code.md): Store code, docs, and assets produced by coding agents in versioned repositories.
- [Distribute Files to Agents](https://docs.tensorlake.ai/filesystems/distribute-files.md): Distribute versioned manuals, skills, configs, and binary tools to agents with read-only mounts.
- [Harbor](https://docs.tensorlake.ai/sandboxes/harbor.md): Run Harbor evaluations and RL rollouts on Tensorlake Sandboxes: fresh isolation per trial, pre-warmed snapshots for expensive environments, and independent test verification.
- [Tool Calls](https://docs.tensorlake.ai/sandboxes/tool-calls.md): Expose Tensorlake sandboxes as tools to your LLM agents, giving models a fresh, isolated execution environment for code, shell, and file operations.
- [Run OpenCode with Tensorlake Sandboxes](https://docs.tensorlake.ai/sandboxes/opencode.md): Route OpenCode's file and shell tools into a Tensorlake sandbox with a single plugin. The model edits, runs, and searches inside an isolated environment instead of on your machine.
- [Run Your Test Suite with Crabbox](https://docs.tensorlake.ai/sandboxes/crabbox.md): Crabbox's Tensorlake provider drops any command into a Firecracker microVM: warm a sandbox, sync your working tree, and run your tests with one command.
- [Run Devin Outposts on Tensorlake Sandboxes](https://docs.tensorlake.ai/sandboxes/devin-outposts.md): Serve Devin Outposts sessions on Tensorlake sandboxes. Devin runs the agent loop; every command, file edit, and repo checkout runs in a Firecracker microVM you control.
- [Run Claude Managed Agents on Tensorlake Sandboxes](https://docs.tensorlake.ai/sandboxes/claude-managed-agents.md): Use Tensorlake sandboxes as the self-hosted execution environment for Claude Managed Agents: Anthropic runs the agent loop, every tool call runs in a sandbox you control.
- [Drive Chrome over CDP](https://docs.tensorlake.ai/sandboxes/chrome-cdp.md): Run Google Chrome inside an ubuntu-vnc sandbox and drive it locally through the Chrome DevTools Protocol over a tunnel.
- [Agentic Swarm Intelligence](https://docs.tensorlake.ai/sandboxes/agentic-swarm-intelligence.md): Orchestrate a swarm of LLM agents running specialized tasks in parallel sandboxes.
- [Agentic Dungeons & Dragons](https://docs.tensorlake.ai/sandboxes/agentic-d&g.md): Build a dynamic D&D-style game where parallel AI agents act as scene writers and a Dungeon Master agent orchestrates the story.
- [RL Training with GSPO](https://docs.tensorlake.ai/sandboxes/gspo-agentic-rl.md): Fine-tune a language model on code generation tasks using Group Sequence Policy Optimization, with TensorLake sandboxes as the reward oracle.
- [Reproducible Environments for RL Rollouts](https://docs.tensorlake.ai/sandboxes/agentic-rl-reproducible-env.md): Use Tensorlake sandboxes to guarantee isolated, deterministic rollouts for reinforcement learning training.
- [Agentic Autoresearch Loop](https://docs.tensorlake.ai/sandboxes/agentic-autoresearch.md): Autonomously improve an ML training script overnight using an LLM agent that proposes code modifications, races them in parallel sandboxes, and hill-climbs toward lower validation loss.
- [CICD & Build Systems](https://docs.tensorlake.ai/sandboxes/cicd-build.md): Execute build steps and run tests in isolated, reproducible environments.
- [Data Analysis](https://docs.tensorlake.ai/sandboxes/data-analysis.md): Perform parallel data analysis and model benchmarking in isolated sandboxes.
- [Use a sandbox as your dev environment](https://docs.tensorlake.ai/sandboxes/remote-dev.md): Get a portable cloud development workstation: SSH in from any machine, idle-suspend when you're not using it, resume by name with state intact.

## Orchestration

- [Introduction](https://docs.tensorlake.ai/applications/introduction.md): Add serverless orchestration to any agent
- [Programming Agents](https://docs.tensorlake.ai/applications/overview.md): Core concepts and common patterns for running agents on Tensorlake
- [Architecture](https://docs.tensorlake.ai/applications/architecture.md): How Tensorlake's Application Runtime runs your code under the hood
- [Building Workflows](https://docs.tensorlake.ai/applications/building-workflows.md): Build multi-step data workflows with parallel execution and optimized resource usage
- [Durable Execution](https://docs.tensorlake.ai/applications/durability.md): Tensorlake automatically persists function outputs so retries and replays skip already-succeeded work, avoiding costly restarts of long-running agent workflows.
- [Crash Recovery](https://docs.tensorlake.ai/applications/crash-recovery.md): How agents survive failures and resume without losing work
- [Futures](https://docs.tensorlake.ai/applications/futures.md): Use Futures to run multiple function calls in parallel to optimize resource usage and reduce latency.
- [Map-Reduce](https://docs.tensorlake.ai/applications/map-reduce.md): Use map-reduce patterns in Tensorlake Applications to parallelize large-scale ETL. Apply a function across items and aggregate the results.
- [Parallel Sub-Agents](https://docs.tensorlake.ai/applications/parallel-sub-agents.md): Fan out work to specialist agents that run in parallel
- [Retries & Rate Limits](https://docs.tensorlake.ai/applications/retries.md): Handle LLM rate limits, transient failures, and structured output validation with durable retries
- [Error handling](https://docs.tensorlake.ai/applications/error-handling.md): How errors propagate in Tensorlake Applications, covering function exceptions, timeouts, retries, and patterns for building resilient agentic workflows.
- [Autoscaling](https://docs.tensorlake.ai/applications/scaling-agents.md): Autoscaling guide for Orchestration endpoints
- [Cron Scheduler](https://docs.tensorlake.ai/applications/cron-scheduler.md): Schedule recurring invocations of your Orchestration endpoints.
- [Observability](https://docs.tensorlake.ai/applications/observability.md): Built-in tracing, execution timelines and monitoring
- [Container Images](https://docs.tensorlake.ai/applications/images.md): Define per-function container images declaratively with Tensorlake's `Image` API. Set the base image, install Python and system packages, and customize per-deploy.
- [Secrets](https://docs.tensorlake.ai/applications/secrets.md): Providing secrets to Tensorlake functions
- [Public Endpoints](https://docs.tensorlake.ai/applications/public-endpoints.md): Expose an application endpoint that callers can invoke without a Tensorlake API key.
- [Async Functions](https://docs.tensorlake.ai/applications/async-functions.md): Use Python async/await with Tensorlake async functions.
- [Timeouts](https://docs.tensorlake.ai/applications/timeouts.md): How function timeouts work and how progress updates reset them
- [Scale-Out & Queuing](https://docs.tensorlake.ai/applications/scale-out-queuing.md): Workflows scale automatically as endpoints are called, with configurable scaling per function
- [Progress Updates](https://docs.tensorlake.ai/applications/guides/streaming-progress.md): Stream real-time progress updates from functions
- [Logging](https://docs.tensorlake.ai/applications/guides/logging.md): Emit logs from Tensorlake applications with `print`, the built-in application logger, or structlog for structured JSON logs that are easier to analyze and visualize.
- [Troubleshooting](https://docs.tensorlake.ai/applications/production/troubleshooting.md): Common issues building Tensorlake applications and how to debug them

## Examples

- [Agent with Tool Calling](https://docs.tensorlake.ai/examples/agentic-applications/agent-with-tools.md): Build a Claude agent that orchestrates complex workflows using tool calls.
- [Deep Research Agent](https://docs.tensorlake.ai/examples/agentic-applications/deep-research.md): Build a multi-agent deep research pipeline with Tensorlake and OpenAI.

## API Reference

- [API Reference Introduction](https://docs.tensorlake.ai/api-reference/v2/introduction.md): Tensorlake v2 API reference entry point.
- [Sandboxes API — Create](https://docs.tensorlake.ai/api-reference/v2/sandboxes/create.md): Create an ephemeral or named sandbox. See sibling endpoints for list / get / update / delete / snapshot / restore / suspend / resume.
- [Sandbox Processes API](https://docs.tensorlake.ai/api-reference/v2/processes/introduction.md): Manage sandbox processes through the sandbox proxy (start, list, signal, stdin/stdout/stderr, follow).
- [Sandbox Files API](https://docs.tensorlake.ai/api-reference/v2/sandbox-files/introduction.md): Read, write, delete, and list files through the sandbox proxy.
- [Sandbox PTY API](https://docs.tensorlake.ai/api-reference/v2/pty/introduction.md): Create and manage interactive PTY sessions through the sandbox proxy.

- [OpenAPI Spec](https://docs.tensorlake.ai/api-reference/openapi.yaml): Raw OpenAPI schema for the Tensorlake APIs, useful for exact endpoint inspection and client generation.

- [llms-full.txt](https://docs.tensorlake.ai/llms-full.txt): Full text of every documentation page — use when you need the complete content, not just the index.