Glasshq

An LLM observability and control platform that traces every agent interaction, LLM call, tool use and chain of thought to detect and classify production failures such as inaccurate responses, broken tool outputs and prompt injections. It is for developers, data and operations teams in B2B companies building and scaling LLM-powered agents. It is delivered as a SaaS platform with an SDK and API and a self-hosted option, using a continuous observe-detect-evaluate-optimize loop based on production data.

Key features

• Trace every LLM call and agent interaction
• Full trace view with tool usage
• Chain of thought step tracking
• Automatic failure detection and classification
• Failure impact measurement on users
• Convert failures to evaluation test cases
• Regression eval suite on production data
• Prompt and parameter optimization
• Daily production digests to team channel
• User counts, cost and latency tracking
• SDK instrumentation with one line
• Decorator tracing for tool calls