Runnable, verified recipes for sending LLM/AI telemetry to an OTLP endpoint — generally, and to IceGate specifically.
Every recipe here is verified: it runs against a local model, exports to a local IceGate, and a shared harness queries the data back over Arrow Flight SQL to assert the telemetry actually landed with the right shape.
| Recipe | Language | Provider | Status | What it proves |
|---|---|---|---|---|
python/openai-ollama |
Python 3.14 | Ollama (OpenAI-compatible) | Verified — make verify PASSED |
Auto-instrumented chat spans, manual invoke_agent/execute_tool nesting, token capture, opt-in content capture with redaction |
typescript/openai-ollama |
Node 26 | Ollama (OpenAI-compatible) | Verified — make verify PASSED (semconv opt-in not honored by the JS instrumentation; see recipe README) |
The same, via the JS instrumentation |
Status is filled in as each recipe lands.
- Ollama with
ollama pull gemma4:12b-mlx - IceGate running locally — clone icegatetech/icegate as a sibling directory and run
make run-docker-core-release
Check everything at once:
make doctor
make verify RECIPE=python/openai-ollama
Read CONVENTIONS.md.
MIT — copy these recipes into your own projects freely.