SafetyMP/corporate-site-harness-llm

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 10, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

SafetyMP/corporate-site-harness-llm is a 7.6 billion parameter Qwen2.5-7B-Instruct model, fine-tuned by SafetyMP for specialized corporate/site harness behavior. It excels at providing local assistance with harness phases, CLI arguments, artifact/gate JSON, and enforcing hard refusals against unsafe shortcuts. This model is optimized for understanding and responding to policy-driven interactions within a corporate site harness context, offering guidance on workspace isolation and preventing misuse of premium models.

Loading preview...

Overview

SafetyMP/corporate-site-harness-llm is a 7.6 billion parameter language model, fine-tuned from unsloth/Qwen2.5-7B-Instruct by SafetyMP. It is specifically designed to act as a corporate/site harness assistant, specializing in policy adherence, corp-harness CLI usage, artifact/gate JSON, and issuing hard refusals for non-compliant requests. The model was trained using Unsloth QLoRA on the SafetyMP/corporate-site-harness-training-data dataset, with adapters merged into the full weights.

Key Capabilities

  • Specialized Harness Guidance: Provides assistance with corporate/site harness phases, workspace isolation, and corp-harness CLI arguments.
  • Policy Enforcement: Trained to enforce policies, including refusing unsafe shortcuts like --actor user or inventing PASS results.
  • Artifact/Gate JSON Understanding: Capable of interpreting and responding to queries related to artifact and gate JSON structures.
  • Refusal of Unsafe Actions: Designed to issue hard refusals for actions that would grant unauthorized approvals or misuse premium models.
  • Offline/Local Use: Intended for local and offline assistance, as well as evaluation and demonstrations against its companion dataset.

Good For

  • Local Harness Assistance: Ideal for developers needing quick, offline help with corp-harness operations and policies.
  • Policy Compliance Checks: Useful for evaluating interactions against defined corporate harness policies and identifying non-compliant requests.
  • Training and Demos: Suitable for evaluating and demonstrating harness-specific scenarios using the validation split of its training dataset.

Limitations

This model is a training specialization artifact and does not replace corp-harness gates or user approval. It is not intended for general software engineering tasks or granting program approval. Its domain is highly specific to corporate/site harness contexts, with limited transferability to general coding or non-harness agent frameworks. Outputs should be treated as advisory, as the model can still be prompted into unsafe compliance under adversarial conditions.

Finetunes
1 models