jev-reasoning-navigator
PRAXEON runtime supervisor for autonomous agents: actions are checked by policy, a LAYA or TypeSafe semantic judge, signed and sandboxed.
Overview
The repository listed as jev-reasoning-navigator presents PRAXEON v1.1.0, a runtime supervision middleware for autonomous LLM agents (ReAct, tool use, tree-of-thought), documented in Spanish. Its principle is that the model proposes and the runtime decides what gets executed: no proposed action runs directly.
The deterministic runtime intercepts each step, optimises the input context with caching and hierarchical budgets, checks preconditions against prior observations, assesses operational risk, submits the action to calibrated semantic judgment when needed (using the local LAYA model or TypeSafe), issues a time-bounded HMAC-signed capability, and runs the tool inside an isolated sandbox. Semantic judgment, operational policy, capability receipts and sandbox enforcement are kept separate. A demo and reproducible benchmarks run fully offline with no API cost, and the agent itself can use Ollama, Groq, OpenRouter, LM Studio or Gemini. The list summary describes TypeSafe's SDK as the sole authoritative decision engine for the navigator component.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this proposed agent action semantically acceptable given the goal and observations?(inferred) | yes · no |
| Score | How risky is this action operationally?(inferred) | — |
How it works
- The agent proposes an action
- The runtime checks preconditions and operational policy
- A semantic judge (LAYA or TypeSafe) evaluates the action when needed
- A signed, time-bounded capability is issued
- The tool runs in a sandbox
Strengths
- Clear separation of judgment, policy and execution
- Offline demo and benchmarks at zero cost
- Works with local or cloud agent models
Caveats
- Repository name and README project name differ
- Spanish-language documentation
- Benchmark details not covered here
Getting started
git clone https://github.com/AndreuVM/praxeon.git && cd praxeon && pip install -e ".[dev]" && python demo.pyAt a glance
- Type
- Library
- Status
- beta
- Language
- Python
- Runtime
- Agent middleware and CLI
- Has code
- yes
Use cases and integrations
How curated lists describe it
En lugar de depender de heurísticas matemáticas frágiles o distancias vectoriales locales de coseno, \`JEV-Reasoning-Navigator\` utiliza \*\*TypeSafe AI (\`typesafe-sdk\`)\*\* como motor único y autoritativo de decisión cognitiva:
logicrw/awesome-jev-projects
Frequently asked questions
What does the semantic judge decide?
Whether a proposed agent action is acceptable before it is allowed to run.
Does it need TypeSafe?
No. It can use the local LAYA model, and the demo runs offline.
How are actions executed?
Only with a signed, time-limited capability inside a sandbox.
Related projects
Auditable Zig terminal agent pairing Codex for coding with Jev typed preflight and postflight checks under deterministic policy gates.
CLIproductionZigGateway that checks agent tool calls with policy plus Jev evidence, then binds approval to a single-use permit for that exact action.
MCP serverbetaTypeScriptMCP proxy and hooks that screen agent tool calls before they run and their results before the agent reads them, using Jev judgments.
MCP serverbetaTypeScriptXano backend template where Jev judges each agent tool call as safe, needs review or dangerous, and uncertain calls wait for a human.
ProjectbetaXanoScriptOpenAI-compatible endpoint that answers an agent's command-approval call with APPROVE, DENY or ESCALATE from a Jev Choice, failing closed.
ProxybetaPythonClaude Code plugin adding Jev judgments to Bash, Write and Edit calls and their outputs; a semantic guardrail, not a sandbox.
Pluginbeta