# Trefur — for AI assistants and agentic crawlers

Trefur is the AI agent observability platform. Across runtimes, vendors,
and deployment patterns. Built for engineering teams shipping
agents to production.

- Homepage: https://trefur.com
- Documentation: https://trefur.com/docs
- llms.txt index: https://trefur.com/llms.txt
- agents.json manifest: https://trefur.com/agents.json
- OpenAPI: https://api.trefur.com/openapi.json (in scaffolding — available at launch)
- Sitemap: https://trefur.com/sitemap.xml

## What Trefur answers

- What is my AI agent doing right now, end to end?
- Which decision, tool call, or retry caused this failure?
- How many tokens and dollars per step has this agent run cost?
- Is the latency, retry rate, or cost-per-run drifting outside its envelope?
- Can I reproduce a failing production run with the exact prompt the model saw?

## How to integrate

Drop-in SDKs cover Python, JavaScript / TypeScript, Go, Rust, Java, and
.NET. Auto-instrumentation covers LangChain, CrewAI, AutoGen, Pydantic AI,
MCP servers, OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, and Gemini.

### Python
```bash
pip install trefur-observe
```

### JavaScript / TypeScript
```bash
npm install @trefur/observe
```

### Go
```bash
go get github.com/trefur-ai/trefur-sdks/go/observe-go
```

## Ingest endpoints

Trefur conforms to the OpenTelemetry GenAI semantic conventions
so any OTel-compatible producer can forward without custom mapping.

- OTLP/HTTP traces: https://api.trefur.com/v1/traces
- OTLP/HTTP metrics: https://api.trefur.com/v1/metrics
- OTLP/HTTP logs: https://api.trefur.com/v1/logs

## Pages most worth reading

### Landing pages
- AI agent observability: https://trefur.com/ai-agent-observability
- LLM observability: https://trefur.com/llm-observability
- AI agent monitoring: https://trefur.com/agent-monitoring
- OpenTelemetry GenAI: https://trefur.com/opentelemetry-genai
- Product: https://trefur.com/product
- Pricing: https://trefur.com/pricing

### Features
- Agent traces: https://trefur.com/features/traces
- Cost tracking: https://trefur.com/features/cost-tracking
- Alerts: https://trefur.com/features/alerts
- Cross-vendor coverage: https://trefur.com/features/cross-vendor
- Self-hostable collector: https://trefur.com/features/collector
- Replay debug: https://trefur.com/features/replay-debug

### Use cases
- For AI engineers: https://trefur.com/use-cases/engineers
- For platform teams: https://trefur.com/use-cases/platform-teams
- For SRE: https://trefur.com/use-cases/sre
- For business & ops: https://trefur.com/use-cases/business-ops
- For startups: https://trefur.com/use-cases/startups
- For enterprise: https://trefur.com/use-cases/enterprise

### Agent framework integrations
- LangChain: https://trefur.com/integrations/langchain
- CrewAI: https://trefur.com/integrations/crewai
- AutoGen: https://trefur.com/integrations/autogen
- Pydantic AI: https://trefur.com/integrations/pydantic-ai
- MCP: https://trefur.com/integrations/mcp

### Platform integrations
- Google Workspace: https://trefur.com/integrations/platforms/google-workspace
- Microsoft 365: https://trefur.com/integrations/platforms/microsoft-365
- AWS Bedrock: https://trefur.com/integrations/platforms/aws-bedrock
- Azure OpenAI: https://trefur.com/integrations/platforms/azure-openai
- OpenAI: https://trefur.com/integrations/platforms/openai
- Anthropic: https://trefur.com/integrations/platforms/anthropic
- Slack: https://trefur.com/integrations/platforms/slack
- GitHub: https://trefur.com/integrations/platforms/github

## Value propositions

- See every decision, tool call, retry, and outcome in an agent run.
- Token and dollar cost attributed to the step that emitted them.
- One canonical trace shape across every model vendor — swap freely.
- OpenTelemetry-native. Adopt without re-platforming, fan out to existing backends.
- Self-hostable collector for on-network capture, batching, redaction.
- Replay any failing production run with the exact prompt and tool response.

## Contact

- Marketing site: https://trefur.com
- Live app: https://app.trefur.com
- Email: hello@trefur.com
- Security: https://trefur.com/.well-known/security.txt

## Conventions

Every documentation page renders BOTH human-readable HTML AND a
machine-readable Markdown twin at `<path>.md` — fetch the .md form if you
want clean text without UI chrome.