Skip to main content
The fastest path to your first captured coding-agent session is Native Hooks. Install the package, run the hook installer, and RCLM handles browser login, local credentials, hook registration, MCP setup, redaction sync, and update checks for you.

Install in two commands

1

Install the package

Install the rclm Python package from PyPI:
This installs the rclm CLI, rclm-hooks-install, rclm-login, rclm-update, hook handlers, and the local MCP server entry point.
2

Run the hook installer

Run the installer once:
The installer opens a browser login flow when credentials are missing, writes credentials and preferences to ~/.reclaimllm/config.json, registers hooks for supported local tools, installs MCP config by default, syncs remote redaction settings, and runs the update flow so the local package and hook configuration are current.
You do not need to run rclm-login or rclm-update separately for the default install path. rclm-hooks-install performs the login flow and runs the update flow automatically.
3

Capture a session

Start a Claude Code, Gemini CLI, Codex CLI, Cursor, GitHub Copilot, or OpenClaw session. RCLM captures the session in the background and uploads it after the session ends.You can check the local spool before upload:
4

Verify in the dashboard

Open reclaimllm.com. Your session should appear within a few seconds of ending, with messages, tool calls, file diffs, model metadata, and token usage.

What the default installer enables

rclm-hooks-install is global by default. It installs hooks into the supported tools it finds and enables the common local helpers that make capture useful without extra setup.

Common install recipes

Reuse an existing API key

Use this when you already created a key and do not want the browser login flow:
To target a non-default backend:

Install for one provider

Use provider flags when you only want to touch one local tool’s config file:
You can combine provider flags:

Install only for the current project

Use --local when the hook registration itself should live in the current project folder instead of your home directory:
This is useful when you want project-scoped hook config checked by folder location. If you want global hooks but only want uploads from selected folders, use --include-folder instead.

Capture only selected folders

Use --include-folder when hooks are global but uploads should happen only for specific project trees:
Repeat it to allow multiple folders:
This writes to redaction.include_folders in ~/.reclaimllm/config.json. When include_folders is non-empty, sessions from every other folder are skipped locally before upload.

Exclude private folders

Use --exclude-folder when most folders are allowed but specific paths should never upload:
Excludes are stored in redaction.exclude_folders. include_folders takes precedence: if an include allowlist exists, RCLM only uploads included folders and ignores excludes for deciding the allowed set. Folder filters accept absolute paths and glob-style folder patterns. RCLM excludes **/.codex/memories/ by default so local Codex memory files are not uploaded as sessions.

Optional feature flags

MCP

MCP is installed by default. Disable it only when you do not want local assistants to access ReclaimLLM memory tools:
MCP provides tools like search_sessions, get_session, and summarize_session to supported local clients. See MCP Server for the tool list.

Claude statusline

The Claude Code statusline is installed by default. Disable it with:
See Statusline for the statusline fields and behavior.

Compression, read cache, and loop breaker

These Claude Code helpers are on by default:
Disable any helper explicitly:
Use shadow mode to measure compression/read-cache/loop-breaker impact without rewriting tool inputs or outputs:
Enable hash dedupe for repeated large tool outputs:
See Context Compression for how these mechanisms reduce token usage.

Data loss prevention

Enable local DLP when you want hook inputs and outputs scrubbed before upload or model exposure:
DLP and redaction settings live under redaction in ~/.reclaimllm/config.json. Remote substitutions managed in the app are synced locally by rclm-update and during install. See Sensitive Data for substitution rules and folder filters.

Context pack

Context pack injects highlights from recent ReclaimLLM sessions in the same project at Claude session start:
It is off by default because it adds a backend lookup to session startup. If credentials are missing or the backend is unavailable, it skips silently.

Handoff advisor

Handoff advisor is on by default. It suggests a focused /compact command and the ReclaimLLM handoff MCP tool when a Claude session grows large. Disable it with:
Default thresholds are 160000 transcript-estimated tokens or 120 tool calls. Customize them in ~/.reclaimllm/config.json:

Config file reference

The installer writes credentials and local preferences to ~/.reclaimllm/config.json. A typical configured file looks like this:
You can edit this file directly for local-only preferences. Re-run rclm-hooks-install after changing install-time choices, or run rclm-update after changing app-managed substitution settings.

Manual login and update commands

The default install path handles login and update automatically. These commands are still available when you want to run those steps separately:
Use rclm-login to refresh local credentials without reinstalling hooks. Use rclm-update to check for package updates, sync redaction settings, and refresh hook configuration after upgrades.

Troubleshooting

Start by checking whether session files are being spooled locally:
If the directory is empty after a completed agent session, re-run rclm-hooks-install and check that it reports hook installation for your provider.
Prefer global hooks plus an include allowlist when you use several local tools but only want uploads from selected repos:
Prefer --local when you want hook config written only inside the current project directory:
Run rclm-update on each machine where hooks are installed. This refreshes the local remote-substitution cache used before upload.
Open ~/.reclaimllm/config.json and confirm server_url, api_key, folder filters, and feature flags are what you expect.

Alternative capture methods

Native Hooks give you the richest local coding-agent data, but other capture methods work independently:

Local API Proxy

Developer machine only: pip install 'rclm[proxy]', run rclm-proxy, then OPENAI_BASE_URL=http://localhost:9800 (or Anthropic/Azure/Gemini equivalents). Uses your provider keys. Not the hosted enterprise gateway.

Browser Extension

Captures ChatGPT, Claude.ai, and Gemini conversations directly in Chrome. Currently in developer mode while awaiting Chrome Web Store review.

Enterprise Gateway

Hosted org proxy at https://proxy.reclaimllm.com/<org-slug>/v1 with team keys (rclm-gw-…). Enterprise-only — do not confuse with rclm-proxy on localhost.
Local proxy ≠ enterprise gateway. rclm-proxy / localhost:9800 is personal capture. proxy.reclaimllm.com is the governed enterprise path. See Capture overview.
Browser Extension and Local API Proxy do not require native hooks. Capture methods produce normalized session records in your dashboard.