5 1 week ago

Local intent compiler for evidence-first AI workflows — benchmarked on the CommerceAI intent registry.

vision tools thinking
ollama run gokeyyyn/company-ai:preview

Applications

Claude Code
Claude Code ollama launch claude --model gokeyyyn/company-ai:preview
OpenCode
OpenCode ollama launch opencode --model gokeyyyn/company-ai:preview
Hermes Agent
Hermes Agent ollama launch hermes --model gokeyyyn/company-ai:preview
OpenClaw
OpenClaw ollama launch openclaw --model gokeyyyn/company-ai:preview

Models

View all →

Readme

Company AI — preview

A local, evidence-first intent compiler for CommerceAI/VibeOperator workflows. It converts natural-language tasks into structured intent decisions while keeping execution, approval and terminal DONE authority outside the model.

Run

ollama run gokeyyyn/company-ai:preview

Reproducibility and benchmark

The public release repository contains the model card, license, deterministic benchmark inputs/results, checksums and an offline verifier:

https://github.com/gokhansahinbas/company-ai

Preview benchmark: 100 of 100 sampled intent-routing cases, 20 of 20 UNKNOWN controls, 0 invalid JSON and 0 execution-authority violations. These figures describe the published benchmark artifact, not a claim of general model quality.

Runtime

The reviewed native Windows executor is open source:

https://github.com/gokhansahinbas/commerceai-pc-relay

Base family: Qwen3.6-35B-A3B · quantization: Q4_K_M · license: Apache-2.0.