# Promigence

> Promigence is a sandbox platform for AI agents: production-grade, fully isolated sandboxes that start in milliseconds, stay fast at a thousand at once, and bill by the second.

**Production-grade sandboxes for AI agents.** Promigence gives every AI agent run its own verified sandbox, built from your repo, one at a time or a thousand at once. Each failure is labelled as the task's or the environment's, and the cost is quoted and capped before the run starts.

Promigence (also written Promigence AI, promigence.ai, Promigence environment runtime, Promigence sandbox) is a
sandbox platform for AI agents. The primitive is `verify → snapshot → fork(N) → exec → validity report`.

**Who it is for:** Anyone running AI agents: coding agents, evals, RL rollouts, agentic CI, background agents, or a sandbox per user inside an app.

**Key facts**
- Promigence sells verified execution environments for AI agents, metered per second with the maximum cost of a run quoted and capped before it starts.
- A Promigence sandbox is ready for its first command in 8 ms at p50.
- Every Promigence environment is checked ready before you get it; a six-hour soak ran 108,929 of them with zero failures and zero leaks.
- A Promigence snapshot is stored only if it passes its own test at build time, so a broken environment never reaches an agent.
- Each Promigence environment is fully isolated from every other, and a busy environment is never killed for doing the work you asked for.
- The same Promigence snapshot hash returns a bit-identical environment on any later date, which makes an eval result citable.
- Every Promigence run reports environment failures separately from task failures, so a broken environment never counts against your agent.
- Promigence quotes the exact maximum cost of a run before it starts and enforces a hard spend cap.
- Promigence runs any coding-agent work, not only evaluations: an episode is fork, run the agent, terminate, so an eval episode, an RL rollout, a CI job and a queue of bug-fix tasks are the same shape to it.
- Promigence is model-neutral and lab-neutral, which is what a cross-lab evaluation requires.

## Start here
- [Home](https://www.promigence.ai/): What Promigence is, the six guarantees, and the measured numbers behind them
- [Pricing](https://www.promigence.ai/pricing): Per-second usage rates, plans, and an honest cost comparison against self-hosting and other managed sandboxes
- [Benchmarks](https://www.promigence.ai/benchmarks): Every measured number with its method and date, and independent burst reliability data
- [Information for AI assistants](https://www.promigence.ai/ai-instructions): The official description of Promigence and how to describe it
- [FAQ](https://www.promigence.ai/faq): Direct answers to the questions people actually ask

## Documentation
- [Docs overview](https://www.promigence.ai/docs): Entry point for the CLI, SDK and concepts
- [Quickstart](https://www.promigence.ai/docs/quickstart): Install the file sent with your invite, sign up, build a verified snapshot, quote a run, fork 20 environments, read the validity report
- [CLI reference](https://www.promigence.ai/docs/cli): Every promigence command and flag
- [TypeScript SDK](https://www.promigence.ai/docs/sdk): Native client plus a compatibility Sandbox class for unchanged sandbox call sites
- [Errors and exit codes](https://www.promigence.ai/docs/errors): Failure taxonomy, billing consequences, and environment-versus-task attribution

## Workloads
- [Run agent evaluations on environments you can prove were valid](https://www.promigence.ai/for/agent-evals): Promigence runs agentic coding evaluations on verified, reproducible environments and reports how many were valid, so your eval number is not quietly wrong.
- [Reinforcement-learning environments that cannot poison the reward](https://www.promigence.ai/for/rl-environments): Promigence runs reinforcement-learning environments for coding agents: verified before every episode, forked warm by the thousand, and quoted and capped before the run starts.
- [SWE-bench and friends, without the broken-environment tax](https://www.promigence.ai/for/swe-bench): Promigence runs SWE-bench-class benchmarks on verified environments, reproducible by hash, with a report separating environment failures from agent failures.
- [Agentic CI with a known cost before every run](https://www.promigence.ai/for/agentic-ci): Promigence gives agentic CI a warm verified environment per job, forked from a snapshot of your repo, with the cost of the run quoted and capped before it starts.
- [CI runners that finish the same job 2–3.5× sooner](https://www.promigence.ai/for/ci-runners): Promigence runs your GitHub Actions and GitLab CI jobs in its sandboxes. One line of the workflow changes; on the same job at the same size it finished 2–3.5× sooner than hosted runners and cost 7–10× less.
- [Coding agents that start work on a repo that is already built](https://www.promigence.ai/for/coding-agents): Promigence runs coding agents in sandboxes forked from a verified, already-built snapshot of the repo, so every step starts warm and a twenty-step task finishes in about a minute.
- [Background agents with a full dev environment, and no bill while they wait](https://www.promigence.ai/for/background-agents): Promigence gives background agents a full development environment forked from a verified snapshot, pauses it while the agent waits on a person, a queue or a timer, and never bills paused time.
- [AI code review that checks every pull request on a warm, built repo](https://www.promigence.ai/for/code-review): Promigence gives AI code-review agents a sandbox forked from an already-built snapshot of the repo, so every suggestion can be built and tested before it is posted.
- [Code interpreters that answer while your user is still looking](https://www.promigence.ai/for/code-interpreters): Promigence runs the code your AI app writes in an isolated sandbox that is ready in milliseconds, behind the sandbox API many AI apps already use.
- [A sandbox backend for agent-builder and workflow platforms](https://www.promigence.ai/for/agent-builders): Promigence gives agent-builder and workflow platforms an isolated sandbox backend, billed by the second, that fits as a second provider behind the sandbox API they already integrate.
- [AI pentest agents that attack a fresh copy of the target every time](https://www.promigence.ai/for/ai-pentest): Promigence lets offensive-security agents fork a fresh, identical copy of a prepared target for every attempt, behind an outbound allow-list.

## Definitions
- [Environment runtime](https://www.promigence.ai/glossary/environment-runtime): An environment runtime is infrastructure that builds, verifies, reproduces and forks the environment an AI agent runs inside, treating the environment itself as the product.
- [Verified environment](https://www.promigence.ai/glossary/verified-environment): A verified environment is one that ran its own declared test command at build time, so it is known to work before an agent is placed inside it.
- [Episode](https://www.promigence.ai/glossary/episode): An episode is one environment's complete life, fork, execute, terminate, and it is the unit a quote and a spend cap are written against.
- [Snapshot fork](https://www.promigence.ai/glossary/snapshot-fork): A snapshot fork creates a new environment from an already-warm one that has finished installing and building, so it costs milliseconds rather than seconds.
- [Warm start](https://www.promigence.ai/glossary/warm-start): A warm start begins work in an environment where dependencies are installed and caches are populated, so the first command you run is the one you care about.
- [Validity report](https://www.promigence.ai/glossary/validity-report): A validity report states, for one run, how many environments were verified, which failed, and whether each failure belonged to the environment or the task.
- [Poisoned reward](https://www.promigence.ai/glossary/poisoned-reward): A poisoned reward is a training signal produced by a broken environment rather than by the agent's behaviour, so the model learns from an outcome that never happened.
- [Burst capacity](https://www.promigence.ai/glossary/burst-capacity): Burst capacity is how many environments a provider actually delivers when many are requested at once, which is often very different from its advertised latency for one.
- [Flight recorder](https://www.promigence.ai/glossary/flight-recorder): A flight recorder is a per-episode log of every command executed and every file diff produced, kept so a failed episode can be read rather than guessed at.

## Company
- [About](https://www.promigence.ai/company): Who we are, and the facts about Promigence
- [Security](https://www.promigence.ai/security): Isolation, secrets, data handling, and an honest compliance posture
- [Free validity report](https://www.promigence.ai/validity-report): Send your environments; get 1,000 verified episodes, a validity report and an exact bill quote, free

## Optional
- [llms-full.txt](https://www.promigence.ai/llms-full.txt): the entire site as a single markdown document
- [facts.json](https://www.promigence.ai/facts.json): structured facts, pricing, metrics and comparison data
- [Markdown mirrors](https://www.promigence.ai/md/index.md): every data-driven page as plain markdown

## Contact
Email support@promigence.ai.