Configure Sensitive Data Detection and Redaction in RCLM

AI sessions frequently contain content that should never leave your machine unreviewed: API keys accidentally pasted into a prompt, database connection strings passed to an agent, or personal information surfaced in a code review. RCLM flags this content automatically on Paid plans and gives every user tools to apply redaction rules before sessions are ever uploaded.

How automatic flagging works

On Paid and Enterprise plans, RCLM checks eligible session and browser-chat records as part of server-side AI summarization after capture and upload. This is different from local hook redaction, which runs before upload. For the check, RCLM builds a bounded view of the conversation: up to 80 user turns, with up to 1,000 characters from each user and assistant side of a turn, up to 15 changed-file paths, and aggregate tool names and counts. The summarization model is instructed to identify:

API keys and plaintext passwords
Email addresses and full names
Local user and file paths
Cloud IAM configuration
Other content that may make the session unsafe to show publicly

The model returns a public-safety flag and a map of detected text to the reason it may be sensitive. RCLM stores that result with the session metadata. This is a heuristic AI review, not a deterministic or comprehensive PII scanner: raw tool-result bodies and full file diffs are not included in this check, and false positives or missed sensitive content are possible.

Automatic flagging happens after the session reaches RCLM and does not redact the stored transcript. To keep known PII or secrets off RCLM’s servers, configure local substitutions or folder capture filters before capture.

Automatic sensitive data detection requires a Paid or Enterprise plan. Free users can review sessions manually before sharing and configure global redaction substitutions for hook capture.

Where to see the outcome

The check result is currently stored as internal session metadata. The web app does not yet show the detected text and reasons, and a sensitive-session marker is not currently available in the session list. The Share Session dialog contains a generic warning state for sessions marked unsafe to show publicly. However, the current session API does not yet return that flag to the web app, so do not rely on the warning as your only review step. Review the transcript before sharing or exporting.

Local hook redaction

Local hook redaction runs on your machine, before sessions are uploaded. When a hook captures a session, the uploader applies your configured substitution rules to the serialized payload — so matched strings are replaced before anything leaves your machine. This is the most effective layer: no sensitive value ever reaches RCLM’s servers if your rules cover it.

How rules are applied

RCLM computes an effective substitution map by combining remote substitutions (managed in the app, synced via rclm-update) and local substitutions (defined directly in your local config file). If the same source string appears in both, the local value takes precedence — giving you machine-specific overrides on top of account-wide defaults. Substitutions use deterministic longest-first string replacement. This prevents a shorter key from replacing text inside a longer matching key before the longer one can match.

Config file structure

Your local redaction config lives at ~/.reclaimllm/config.json:

{
  "redaction": {
    "enabled": true,
    "remote_substitutions": {},
    "local_substitutions": {
      "my-secret-value": "[REDACTED]",
      "postgres://user:pass@host/db": "[DB_CONNECTION]"
    },
    "include_folders": [
      "/Users/you/work"
    ],
    "exclude_folders": [
      "**/.codex/memories/",
      "/Users/you/projects/private-client",
      "/home/you/work/internal-infra"
    ],
    "last_sync": "2026-04-27T10:00:00Z"
  }
}

Remote vs. local substitutions

Remote substitutions

Managed under Global Substitutions on the Settings page. Synced to your machine when you run rclm-update. Use these for account-wide secrets, shared team tokens, or any value that should be redacted on every machine where you use RCLM.

Local substitutions

Added directly to ~/.reclaimllm/config.json under redaction.local_substitutions. Use these for machine-specific values: local usernames, absolute paths, workstation-only credentials, or private service aliases that should not become account-wide rules.

Folder capture filters

Folder filters are stricter than substitutions — instead of replacing sensitive strings, they decide whether the entire session is uploaded. Use redaction.include_folders when you want RCLM to capture only specific project trees. Set it during installation with rclm-hooks-install --include-folder=/path/to/project, or edit ~/.reclaimllm/config.json directly. When this list is non-empty, a session is uploaded only if its working directory or transcript path matches one of the included folders. Sessions from every other folder are skipped locally. Use redaction.exclude_folders for paths that should never upload. Set it during installation with rclm-hooks-install --exclude-folder=/path/to/private, or edit the same config file directly. Exclusions apply when include_folders is empty. If include_folders is set, the include allowlist supersedes exclude_folders so matching included projects are captured. 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.

Folder filters apply to the machine where they are configured. If you use RCLM on multiple machines, add the same include or exclude entries to each machine’s local config file.

Keeping redaction rules up to date

Configure shared substitutions in the app

Open Settings, scroll to Global Substitutions, and add the rules you want applied on every machine. These become your remote substitutions.

Run rclm-update on each machine

After adding, editing, or removing shared substitutions, run the following command on every machine where hooks are installed:

rclm-update

This refreshes the local hook redaction cache from the app’s current remote substitutions.

Add local-only substitutions

Open ~/.reclaimllm/config.json and add any machine-specific values under redaction.local_substitutions. These are never pushed to the app.

Add folder capture filters

During installation, pass --include-folder=/path/to/project for project allowlists and --exclude-folder=/path/to/private for deny-lists. You can also edit redaction.include_folders and redaction.exclude_folders directly in ~/.reclaimllm/config.json later.

Verify with a test capture

Run a short session in a tool with hooks installed and confirm the redacted values appear as their replacements in the RCLM session detail view.

Run rclm-update after every substitution change in the app. Hooks use a locally cached copy of your remote rules, and the cache does not refresh automatically — only when you run rclm-update.

Review the full transcript before creating a share link or export. Automatic flagging is an advisory, post-upload check and does not redact the transcript or block sharing. A generic warning state exists in the Share Session dialog, but the detector flag is not yet wired through the current session API. Detected text and reasons are not currently shown in the web app, and flagged sessions are not currently marked in the session list.

Enterprise DLP

Enterprise plans include Data Loss Prevention (DLP) scanning and org-wide policy enforcement.

Org-wide DLP rules

Admins configure DLP policies that apply across all org members’ captured sessions — not just per-user substitutions. Coming soon.

Sensitive data alerts

Admins receive alerts via Slack, PagerDuty, or webhook when sessions matching DLP rules are captured. Coming soon.

Enterprise orgs can Require DLP from the Coverage tab so local --no-dlp is overridden at runtime (with optional time-boxed exceptions). Hook removal and DLP-off posture surface as AD-010 / AD-011 — see Coverage Alerts.

Plan comparison

Capability	Free	Paid	Enterprise
Automatic sensitive data detection	—	✅	✅
Hook redaction (local substitutions)	✅	✅	✅
Remote substitution sync (`rclm-update`)	✅	✅	✅
Folder include/exclude filters	✅	✅	✅
Sensitive content warning before sharing	Manual review	Not yet wired	Not yet wired
Auto-redaction before storage	Manual only	Hook rules	Org-wide policies (soon)
DLP scanning + policy enforcement	—	—	Partial now; org enforcement soon
Sensitive data alerts (Slack / webhook)	—	—	Coming soon

Get Started

Capture

User Portal

Privacy & Security

Enterprise

Plans & Billing

Configure Sensitive Data Detection and Redaction in RCLM

How automatic flagging works

Where to see the outcome

Local hook redaction

How rules are applied

Config file structure

Remote vs. local substitutions

Remote substitutions

Local substitutions

Folder capture filters

Keeping redaction rules up to date

Enterprise DLP

Org-wide DLP rules

Sensitive data alerts

Plan comparison

​How automatic flagging works

​Where to see the outcome

​Local hook redaction

​How rules are applied

​Config file structure

​Remote vs. local substitutions

Remote substitutions

Local substitutions

​Folder capture filters

​Keeping redaction rules up to date

​Review before sharing

​Enterprise DLP

Org-wide DLP rules

Sensitive data alerts

​Plan comparison

How automatic flagging works

Where to see the outcome

Local hook redaction

How rules are applied

Config file structure

Remote vs. local substitutions

Folder capture filters

Keeping redaction rules up to date

Review before sharing

Enterprise DLP

Plan comparison