System One Harness
Turns a decision model into an agent loop: it compiles the environment's finite action space into typed questions, gates each step by confidence, and records the trace.
- GitHub stars
- 101 as of 2026-09-22
- Forks
- 13
- Language
- Python
- License
- Apache-2.0
- Created
- 2026-09-19
- Last updated on GitHub
- 2026-09-21

What Jev does in System One Harness
The project describes itself this way: Turns a decision model into an agent loop: it compiles the environment's finite action space into typed questions, gates each step by confidence, and records the trace. In the Community additions category, Jev typically decides a single well-defined judgment, such as a label, a score or a yes/no check.
Verification
Verified. We confirmed this repository and its Jev relevance through the GitHub API and the source lists below. A specific file link has not been recorded yet.
How Community additions projects use Jev
Projects added from community lists and direct submissions. Many are small, single-purpose experiments, which makes them some of the easiest places to see how a Jev question is framed in real code.
Related Jev projects in Community additions
lurkgetanyapi-com/lurkSelf-hostable Reddit buyer-intent finder that uses Jev to judge every post and comment a scan reads.★ 97
TypeScript
Supercovsupercorp-ai/supercovCode quality for coding agents: Jev answers twelve `Noul` properties per source file so the agent knows what to fix first.★ 105
Rust
Jev Chatw3cj/jev-chatChat-style command bar where Jev picks the tool, arguments, and reply type, and code builds every reply from tool data.★ 91
TypeScript- jev-usesavka777/jev-usemacOS computer use by voice or typing that reads the screen through the Accessibility tree rather than screenshots. Key stored in the Keychain.★ 90
Swift jevchatkyle-pena-nlp/jevchatTurns a decision model into a chat model by asking which symbol comes next, then sampling from the returned distribution.★ 88
Python- webctldorkitude/webctlSearch CLI for agents: results from up to three backends are scored by Jev against your query and an explicit `--goal`, and only the relevant ones reach the agent's context. Its benchmark excludes an arm it could not observe.★ 119
Go