Skip to main content
Lynx records how an AI agent moves from a user request to LLM calls, tool calls, policy decisions, retries, and a final outcome. Use the TypeScript SDK or connect any runtime over HTTP.
Lynx is currently a pilot-stage project. APIs and behavior may change before the general release. To join the pilot or evaluate Lynx for your service, contact partners@hvnn.me.

TypeScript quickstart

Install @lynxops/sdk and record your first agent Run.

OpenAPI quickstart

Build an integration or SDK in Python, Go, Java, Rust, or another language.

​
Choose an integration

Use the TypeScript SDK when you want built-in async context, LLM and tool wrappers, local firewalls, batching, retries, and runtime configuration caching. Use the OpenAPI directly when:
  • Your service uses another language or runtime.
  • You are building a community SDK.
  • You already have an event queue and delivery system.
  • You need explicit control over local storage and policy execution.
Both integrations use the same Run, Session, Event, prompt version, policy, and issue record model.

​
Core flow

Lynx can show inputs, execution steps, LLM and tool calls, errors, retries, latency, token usage, cost, policy decisions, and related prompt versions in one Run.

​
Production safety

Event delivery should run in the background. A Lynx API outage should not fail your customer request. Use bounded queues, short timeouts, selective retries, and a circuit breaker. Firewall checks run locally before the tool executes. High-risk actions can use fail-closed or approval behavior without depending on a network call.
Prompts, tool arguments, model responses, logs, and payloads can contain sensitive data. Mask secrets before recording them and use metadata-only capture when raw content is not required.

​
Start building