Our take
Newless than six months old- Activity
- 35
- Community
- 29
- Issues
- 21
- Pull requests
- 33
+2 stars in the last 30 days
- Slower to merge pull requests than 89% of the projects we track
This is a useful, unusually security-focused shortlist, though its documented comparisons should guide further evaluation rather than stand in for security testing.
Good fit if
- You need to compare how coding-agent sandboxes handle isolation, egress, and secrets rather than just execution speed.
- You want to consider both managed and self-hosted options across several isolation models.
- You value traceable claims and explicit unknowns when documentation does not establish a property.
Look elsewhere if
- You need a sandbox implementation to install and run; this repository is a directory, not a runtime.
- You need penetration-test results or independently validated security guarantees; the project says it reviews documentation, not implementations.
- You need comparisons centered on price, startup latency, throughput, SDK ergonomics, or language coverage, which the list deliberately omits.
All health signals
| Last commit | 2026-11-02 (4 days ago) |
|---|---|
| Commits, last 90 days | 12 |
| Releases, last 12 months | 0 |
| Contributors | 6 (top contributor: 85% of commits) |
| Issues closed, last 90 days | 0 |
| Pull requests merged, last 90 days | 9 (typically merged in 5 days) |
| Project age | 4 months |
Checked on 2026-11-07 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
28Forks
20Change since Nov 14, 2026: 0 stars·0 forks
Overview
awesome-ai-coding-sandboxes is a curated comparison of isolation solutions for code run by autonomous AI coding agents, including Claude Code, Codex, and OpenHands. It focuses on the security boundary and what agent code can reach beyond it, especially network egress and secrets, rather than startup speed.
The project is an actively maintained directory, not a sandbox runtime. Its matrix covers self-hosted and managed options, from microVMs and full VMs to containers and gVisor, and marks undocumented properties as unknown instead of guessing. Each matrix entry is intended to trace back to the project’s own documentation or source.
Key Features
- Compares isolation tiers, including microVMs, full VMs, gVisor, containers, and process sandboxes.
- Records whether outbound network access is denied by default, configurable, open by default, or delegated.
- Distinguishes brokered secrets, kept outside the sandbox, from secrets injected into its environment.
- Notes deployment model and workspace state, including self-hosting, managed services, and persistent or ephemeral workspaces.
- Groups additional projects by sandbox approach and lists underlying isolation building blocks such as Firecracker, Cloud Hypervisor, Kata Containers, and Bubblewrap.
- Explains the limits of the comparison: documented claims are not penetration-test results, and deployment choices can weaken a sandbox’s boundary.
Use Cases
- Security teams can use the matrix to shortlist agent execution environments with both strong isolation and restricted egress.
- Developers choosing a hosted or self-hosted sandbox can compare deployment options, workspace persistence, and secrets handling in one place.
- Platform engineers can identify projects that fit a particular isolation approach, such as microVMs for a stronger kernel boundary or gVisor for user-space-kernel isolation.
- Contributors and maintainers can use the source-linked comparison to check claims or propose corrections when documentation changes.
What you need
Detected in the repository
- Python (main language on GitHub)
- Automated checks on GitHub Actions
License in plain words
CC0-1.0public-domain-like
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: nothing
- Share your changes: no
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
Open the comparison matrix and shortlist candidates by isolation, egress control, and secrets handling. Read each candidate’s own documentation before adopting it. See the README for the full list and contribution guidance.
Alternatives
- agent-sandbox: Provides a local sandbox for running coding agents, rather than a directory for comparing sandbox options.
- n8n-sandbox-service: Provisions managed Docker/Sysbox or Firecracker sandboxes through a central API, rather than cataloging options for comparison.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| awesome-ai-coding-sandboxes | Python | CC0-1.0 | 28 | New |
| agent-sandbox | Python | MIT | 211 | Active |
| n8n-sandbox-service | Go | – | 33 | Active |
Considerations
The project is new but active: it has recent commits and merged pull requests, with contributions concentrated mostly in one person. It has no release history yet, and the measured health data shows no issues closed recently, so the repository’s long-term maintenance pattern is still emerging. The matrix reflects documented behavior, not implementation audits; a listed control may also depend on configuration, and mounting host files or misconfiguring a broker can undermine a boundary. The repository does not provide a runtime, installation command, or tests for adopting one of the listed sandboxes.
Found this useful?
Share it with someone who would like awesome-ai-coding-sandboxes.