Back to directory

Dormice

A self-hosted, E2B-compatible sandbox platform for AI agents where sandboxes persist indefinitely and cost nothing while idle, instead of being disposable cloud sandboxes billed by the second.

Dormice

Dormice is a self-hosted sandbox platform for AI agents, described as the SQLite of agent sandboxes: rather than the disposable, per-second-billed sandboxes typical of cloud sandbox platforms, Dormice runs on a machine the user already pays for and keeps sandboxes permanent, getting cheaper to keep the longer they sit idle. Its core operation, acquiring a sandbox by a user key, is idempotent: the same key always returns to the same sandbox regardless of its current state, waking, starting, or restoring it as needed.

An idle sandbox cools down on its own through active, frozen, stopped, and archived states, one step at a time, and any later acquire brings it back; on measured hardware, freezing an idle sandbox holding 1 GiB of memory reduces it to about 5 MiB resident, and waking it again takes about 50 milliseconds. It deploys as a single daemon backed by one SQLite ledger, needing no Kubernetes or external database, and the official e2b SDK works against it by changing two URLs.

The project is in early development: the daemon, SDK, CLI, web console, and Docker-plus-gVisor executor work end to end but are not yet production-ready.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.