celesto
Secure, persistent VM sandboxes for AI agents, with a PR-review example that compares a general model with Jev on candidate findings.
Overview
Celesto gives AI agents their own computer for running code, browsing and using desktop apps. Each sandbox is a lightweight virtual machine that starts in about 500 ms, can keep files and state between sessions, and runs locally during development or in Celesto Cloud, providing a stronger boundary for untrusted code than a local process. OpenMuse, an open-source computer coworker built on it, browses in a disposable Linux desktop while you watch and approve.
Per the list summaries, a Celesto PR-review example prepares sandbox checks and pairs a general model with Jev to double-check candidate findings on the same diffs. The README excerpt does not describe that example in detail.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this candidate PR-review finding valid?(inferred) | yes · no |
How it works
- Run checks for a PR in a Celesto sandbox
- A general model proposes candidate findings
- Jev double-checks each finding
- Compare results
Strengths
- Isolated VMs for untrusted code
- Example contrasts LLM and decision-model review
Caveats
- Jev example not described in the README excerpt
- Jev is peripheral to the main product
Getting started
See the Quickstart and Python API in the README.At a glance
- Type
- Project
- Status
- production
- Language
- Python
- Runtime
- VM sandboxes (local or cloud)
- Has code
- yes
- Has tests
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
A Celesto PR-review example prepares sandbox checks and compares a general model with Jev on candidate findings.
logicrw/awesome-jev-projects
Frequently asked questions
Where does Jev appear?
In a PR-review example that double-checks findings, per the listing.
What is Celesto?
Fast-starting VM sandboxes for AI agents.
Is it open source?
Yes, under Apache-2.0.
Related projects
Oh My Pi skill for method-first bulk classification: decide when to use Jev, then pilot and freeze a rubric for judge_batch.
Agent skillbetaPythonUnix-style tools where Jev learns a repo's decision norms and judges each node of a code lead as a bug, lead or clean.
ProjectbetaJavaScript'Vitest for LLM outputs': CLI running YAML/JSON eval suites on Jev for RAG faithfulness, hallucination and agent safety in CI.
CLIprototypeTypeScriptLocal PR reviewer: deterministic diff handling plus Jev or Laya judgments of severity, category and true-positive per hunk.
ProjectprototypePython- jev-harness★ 30
Research proposal-review contract: an LLM proposes one action, Jev answers four narrow questions, code produces evidence, applies nothing.
LibraryresearchTypeScript Checks whether each cited paper supports its citing sentence: code fetches papers, Claude proves a quote, Jev scores it, a human decides.
Serviceproduction