Skip to main content
ReclaimLLM captures AI interactions wherever they happen: inside coding agents, browser chat interfaces, local API calls, and org-level gateway traffic. No matter which path you use, every session lands in the same normalized record — searchable, browsable, and exportable from a single dashboard. You can start with just one method and layer on the others as your needs grow.

​
Five capture methods

Native Hooks

The richest capture path. Hooks into Claude Code, Gemini CLI, Codex CLI, Cursor, and GitHub Copilot to record messages, tool calls, file diffs, and token usage.

Browser Extension

Capture ChatGPT, Claude.ai, and Gemini conversations in the background — no API keys or code changes required.

OpenTelemetry

Export traces from existing OpenAI, Anthropic, Gemini, Bedrock, and other LLM API integrations without proxying provider traffic.

Local API Proxy

Local rclm-proxy on localhost:9800 that intercepts requests with your own provider keys. Not the hosted Enterprise Gateway.

Enterprise Gateway

Hosted proxy.reclaimllm.com/<org-slug>/v1 for governed team traffic. Admins issue scoped rclm-gw-… keys and review usage centrally.

​
Comparison

Local proxy ≠ enterprise gateway. Do not point SDKs at proxy.reclaimllm.com when you mean rclm-proxy on localhost, and do not use a personal provider key as if it were a team gateway key. Details: Local API Proxy · Enterprise Gateway.
The free tier includes 200 hooks-captured sessions. Browser extension and local API proxy sessions are unlimited on all plans. OpenTelemetry ingestion and the hosted Enterprise Gateway are Enterprise-only.

​
The normalized session model

Every capture method — regardless of how rich the source data is — produces the same session shape. That consistency is what lets browser chats, agent traces, local API calls, and enterprise gateway requests live together in one dashboard.

Messages

Prompts, replies, and conversation turns from every source.

Provider context

Provider name, model, timestamps, token usage, and cost metadata.

Tool activity

Tool calls and their results, when the source supports them (hooks, proxy, and instrumented OTLP traces).

File diffs

Before-and-after file snapshots for every file an agent created or modified — hooks only.

Session metadata

Auto-generated titles, descriptions, and tags that make sessions immediately searchable.

Sensitive content flags

Credentials, tokens, and other risky content highlighted before storage.
When session encryption is enabled (Paid and Enterprise plans), full session blobs are encrypted at rest. Metadata — titles, tags, token counts, model names — stays unencrypted so search, stats, and summaries remain fast.

​
Choosing where to start

The right first step is the one that captures your highest-volume AI usage with the least friction. You can always add more paths later — they don’t interfere with each other.
  • You use Claude Code, Gemini CLI, Codex CLI, Cursor, or GitHub Copilot daily → Start with Native Hooks
  • You mostly chat in the browser → Start with the Browser Extension
  • Your application already calls LLM provider APIs → Start with OpenTelemetry
  • You build apps that call LLM APIs → Start with the Local API Proxy
  • You manage LLM access for a company → Start with the Enterprise Gateway
  • You want the fullest picture → Use all five paths together