Skip to main content
Belvedir is an observability and auto-improvement platform for LLM agents. Install the SDK, wrap your agent runs in sessions, and Belvedir segments the traces into tasks, groups similar work together, and runs improvement loops — prompt evolution, memory harness updates, and LoRA fine-tuning — against your production traffic.

​
Where to start

  • Quickstart — get traces flowing in under 5 minutes with the Node SDK.
  • Python SDK — same concepts, context managers instead of callbacks.
  • How It Works — the six-stage pipeline: instrument, collect, segment, group, clean, optimize.

​
What you can do

  • Trace every LLM and tool call from OpenAI, Anthropic, Cohere, Bedrock, Vertex AI, and more.
  • Segment sessions into tasks automatically and gather similar tasks into groups.
  • Optimize with prompt evolution (GEPA), memory harness updates, and LoRA fine-tuning that open PRs against your GitHub repo.
  • Benchmark any agent harness in an isolated CPU or GPU sandbox against SWE-bench, τ-bench, GAIA, and more.
  • Train open models on your successful production traffic and download the LoRA adapter.

​
Supported runtimes

The Node SDK (@fractalresearch/loop) and Python SDK (fractal-loop) both auto-instrument the major LLM providers. Local models via Ollama and hosted OpenAI-compatible endpoints (Together, Fireworks, Groq, OpenRouter, vLLM) work out of the box — see Hosted Inference.