Jev Ecosystem: OpenJev, NanoJev, jev-router & Awesome Jev
The Jev ecosystem now includes the official TypeSafe AI model, independent model research, agent routers, open-source applications and several Awesome Jev directories. Similar names do not mean the projects share code, weights, maintainers or official status.
This guide maps the terms people commonly search for—OpenJev, NanoJev, jev-router, awesome-jev, Jev AI and Jev model—to their actual projects and intended use.
AutoJev is an independent integration layer. Community projects listed here are linked for discovery and are not endorsements. Check each repository's current README, license, security model and maintenance status before installing it.
Official Jev AI and the Jev model
Jev is TypeSafe AI's System One decision model. It receives shared state plus typed Choice, Noul or Score questions, then returns bounded answers and probabilities instead of generating an essay.
The official model and API are distinct from community implementations that reproduce a similar interface or experiment with open backbones. Start with the TypeSafe Jev guide for the vendor and model relationship, the Jev model guide for request and response concepts, and the official TypeSafe documentation for the provider's current contract.
What is OpenJev?
OpenJev is an ambiguous community label rather than one official model. At least two independent repositories currently use or previously used that name:
- razorback16/openjev describes an open, Jev-compatible System One decision server built around DiffusionGemma. Its README notes provisional vLLM dependencies and tells users to evaluate answer quality on their own tasks.
- TheoLeeCJ/SemIf, originally published as
openjev, studies semantic decisions from open models on local hardware. The project explicitly says it does not reproduce Jev's undisclosed model or training.
Neither repository is TypeSafe AI's Jev weights. Treat “OpenJev” as a family of independent compatibility and research efforts, and verify which repository a search result or discussion means.
What is NanoJev?
NanoJev is an independent research project built around a small Qwen backbone, typed decision heads and a training/evaluation pipeline. Its repository includes maze and Snake controllers, browser replays, checkpoints and datasets.
NanoJev is useful for studying how a local model can score dynamic candidates and how deterministic game code can consume those scores. It is not an official Jev release, and its recorded benchmark or game results should not be generalized to unrelated production workloads without new evaluation.
What is jev-router?
The name jev-router appears on several projects because routing is a natural fit for bounded Choice decisions. A router supplies an allowed candidate list, asks Jev to select one, then applies deterministic constraints and fallbacks.
gargpratyush/jev-router routes fresh Claude Code and OpenAI Codex turns across model tiers while preserving the host CLI's native tools, sessions and permissions. BillionsBobby/JevRouter is a separate router for models, tools and subagents. Other projects use the term for HTTP routing, skills or LLM gateways.
These packages are independent community tools, not a single standard Jev Router product. For the architecture, safeguards and an AutoJev request example, read Jev Router and model routing.
What is Awesome Jev?
Awesome Jev usually refers to a community-maintained directory of Jev projects. Multiple lists exist, including:
- hellogumbo/awesome-jev, a categorized directory covering official resources, integrations, computer use, games, research and applications.
- yibie/awesome-jev, a source-backed collection organized by use-case categories.
- awesomejev.com and awesomejev.cc, searchable directory interfaces maintained by their respective communities.
Directories help discover projects, but inclusion is not a security review or performance guarantee. Follow the source repository, inspect recent commits and reproduce claims before adopting a project.
Jev ecosystem map
| Search term | What it usually means | Official TypeSafe Jev? |
|---|---|---|
| Jev AI | The decision-model concept and TypeSafe AI product | Yes, when linked to TypeSafe |
| Jev model | The System One model that answers typed questions | Yes, when using the provider model |
| OpenJev | Independent open-model or compatible-server experiments | No |
| NanoJev | Independent small-model research and game experiments | No |
| jev-router | One of several community routing tools or the routing pattern | Usually no |
| awesome-jev | Community directories of Jev links and projects | No |
| AutoJev | Independent REST, MCP and Skills integration layer | No |
The distinction matters for credentials, privacy, licenses and expected behavior. An official SDK can still call a hosted provider, while a repository with “open” in its name may require local weights, a patched runtime or a separate upstream model license.
Popular Jev application clusters
Community projects tend to cluster around decisions that can be enumerated before inference:
- Model, tool and skill routing: choose among approved candidates and retain an escalation path.
- Browser and computer use: choose a bounded operation and a grounded target from DOM, OCR or accessibility state. See Jev computer use.
- Games and simulations: select legal moves from structured world state while deterministic code owns the rules. See Jev games.
- Guardrails and verification: classify tool risk, research support or completion evidence.
- Local decision-model research: compare open backbones, scoring heads, probability calibration and controller design.
How to evaluate a Jev community project
- Identify the actual provider. Determine whether it calls TypeSafe, a gateway, an open model or a local compatibility server.
- Inspect the decision contract. Confirm how Choice, Noul and Score are represented and whether candidates are truly bounded.
- Check secrets and data flow. Know which state leaves the machine and where API keys are stored.
- Reproduce measurements. Treat latency, cost and accuracy numbers as project-specific until you can run the same evaluation.
- Keep authority outside the model. A valid answer does not grant permission to click, spend, publish, delete or change production state.
- Review licenses separately. Repository code, base-model weights, datasets and checkpoints can have different terms.
To build against hosted Jev, continue with the Jev API guide, Jev MCP server, or AutoJev playground. For broader patterns, browse all Jev use cases.