XY Xiaoyuan 14B

Local-first · safety-gated · 14B

A 14B agent that shows its gates.

Xiaoyuan 14B is not a new foundation model. It is an agent architecture around a quantized 14B open-weight model: deterministic scenario routing, safety gates before generation, task constraint blocks, and output repair after it. Built for a 9 GB, 4k-context, local-first world.

Public project surface · live endpoint not claimed
31/31local turns completed
45/45automated regression tests
89/100local strict self-score*
~9 GBQ4_K_M local runtime

*Local self-evaluation, not an official competition score or leaderboard rank.

The failure path is the product.

Most agent demos optimize for the happy path. This one makes the critical path deterministic: stop unsafe work before generation, put emergency action first, keep structured output structured, and repair missing fields before the user sees them.

Scenario router

Crisis, health, safety, tool, decision, multi-domain, self-review, general. Priority is explicit.

Deterministic gate

High-confidence cases short-circuit with zero model latency; the label is auditable.

Task constraint block

A compact, task-specific rule block instead of a generic constitution on every turn.

Local model call

Qwen2.5 14B Instruct, Q4_K_M, Ollama, 4k working context, no cloud dependency.

Output post-check

Repair missing emergency-first action, unsafe framing, format drift, and prompt echo.

What it refuses to hide

  • It is an architecture around Qwen2.5 14B, not a new model.
  • It has no public always-on inference endpoint.
  • Local evidence is not an official benchmark run.
  • A human click-through of the production UI is still a release gate.
  • It does not replace a clinician, lawyer, emergency responder, or adviser.

Evidence, with the limits attached.

The project records local results because they are reproducible. It does not turn them into an official score.

CheckResultScope
Ten-task regression10/10 · 31/31 turnslocal Qwen2.5 14B, optimized mode
Automated regression45/45 tests OKrouting, gates, post-check
Local strict self-score89/100local rubric; not official
Safety-critical itemsQ4 / Q6 / Q10 passedlocal safety rubric
Repeatability3/3 same decisiondeterministic gate path
Public live endpointNot claimedlocal-only deployment

Full evidence note: docs/evidence.md ↗

Where it is being submitted.

Only verified states are listed. “Prepared” is not “submitted”.

TargetWindowStateFit
Create AI Agents Championship · DubaiRegister by Jan 2027Prepared · personal fields neededBest Personal Agent
Open Agentic Web Challenge2026-11-12 → 2027-01-11Public surface ready · Devpost pendingAgent-readable project site
SEC Minimalist Edge AI Challenge 2026ClosedResults published2026 not eligible; 2027 TBD
Kaggle Gemma 4 Developer Agent2026-12-02Not eligible as-isRequires Gemma 4 base
Neo4j Relate Hackathon2026-12-11New work requiredRelate memory graph branch
HackerRank Orchestrate2026-12-12Live build onlyIndividual 24h event

Full matrix: docs/submission-matrix.md ↗

Machine-readable by design.

Another agent should not have to scrape a marketing page to learn what this system does and where it stops. These files are the discovery surface.

GET /agent-card.json GET /llms.txt GET /openapi.json agent "Xiaoyuan 14B" protocol "0.3.0" skills routing, gates, constraints, post-check live_endpoint false

Skills advertised in the card

Loading agent card…

The card is a discovery document, not a promise of a hosted model. A live A2A endpoint is a separate deployment step.