Frequently asked questions

How do I install FORG?

Run: curl -fsSL https://forg.pro/install | sh — then forg activate <your-license-key>. The agent auto-detects your AI tools and wires adapters automatically. See forg.pro/docs/installation for the full guide.

Which AI tools does FORG support?

Claude Code, Cursor, Windsurf, GitHub Copilot (via MCP adapter), and any editor with an MCP-compatible interface. FORG auto-detects installed AI coding tools after activation — no manual adapter wiring required.

Can I switch plans?

Yes. Upgrades take effect immediately. Downgrades take effect at the next billing cycle. Data above the retention limit for the new plan is not deleted immediately — you have 30 days to export before it ages out.

Do you offer nonprofit or open source discounts?

Yes. See our startup, education, and open source program pages for details.

What's the difference between async and gateway mode?

Async mode (default): rules evaluate after the fact. Signals are processed within seconds of a session event, but the actual LLM call is not blocked. Gateway mode: rules evaluate before the LLM call. Blocking rules prevent the call from reaching the provider — zero cost for blocked calls. Gateway mode has P50 3.2ms overhead.

What happens if my budget rule fires?

Depends on the rule action: alert-only logs the event and sends a notification but doesn't block. block prevents new LLM calls from the affected scope until the window resets. redirect sends calls to a fallback model instead of blocking.

Can I test a rule before enabling it?

Yes. Set mode: dry_run in your rule configuration. Dry-run rules evaluate against incoming signals and log matches, but never take enforcement action. Review dry-run matches in the Rules → Dry Run tab before enabling.

How many rules can I have?

Developer: 20. Professional and above: unlimited. Business and Enterprise add organization-wide governance, audit controls, and custom policy administration.

Does FORG see my prompts or code?

No. FORG signal schema v3 captures metadata only: token counts, latency, model, cost, and dimension tags. Prompt text, completion content, and file contents are never collected, transmitted, or stored. This is enforced at the binary level.

What is k-anonymity and why does FORG enforce it?

k-anonymity ensures that no individual can be singled out from aggregated data. FORG enforces a minimum cohort size of k ≥ 5: if a query would reveal data about fewer than 5 users, the result is suppressed. This applies to all dashboard views, API responses, and FORG Atlas answers.

Can I delete my data?

Yes. Go to Settings → Data → Delete account data. You can also export all your signal data before deletion. Account deletion and data export requests are processed within 30 days per GDPR/CCPA requirements.

Where is my data stored?

By default, signal data is stored in Supabase (US East). Business+ customers can elect EU data residency (eu-central-1, Frankfurt). License and identity data is stored in Cloudflare D1 (US-resident, by design).

Why is my cache hit rate showing as 0%?

Cache hit rates depend on how your AI tool constructs requests. Claude Code caches system prompts across sessions; short or one-off sessions typically show 0%. Check multi-turn Claude Code sessions with repeated system context for expected cache activity.

What is FORG Atlas?

FORG Atlas is a natural language query interface for your usage data. Ask questions like 'Which developer spent the most last week?' or 'What is our cache hit rate trend over 30 days?' in plain English. Available on Business+ plans.

The dashboard shows data but my team members don't see it.

Each user must install the FORG agent and activate with their own license key. A seat on the Team/Business plan grants a user dashboard access, but they still need the agent running locally to emit signals.