Documentation Index
Fetch the complete documentation index at: https://docs.complyhat.ai/llms.txt
Use this file to discover all available pages before exploring further.
Annual discount adjusted to 10%.
- Solo annual now 4,790/yr at 20% off). Team annual now 23,990/yr).
- New Stripe price IDs live; legacy 20%-off prices archived.
- Dashboard footer unified with the landing footer. Pricing card simplified to 3 bullets per tier.
- Landing feature tabs reshaped around what ships: Reports, Compliance Wiki, MCP Tools.
Per-MCP-install seat pricing.
- Pricing pivoted from per-AI-model to per-MCP-install seats. Solo 2,499/mo (up to 5 seats). Enterprise custom, priced by seat count.
organizations.seat_limitreplacesmodel_limit.billing_cadencecolumn tracks monthly vs annual.- AI models are uncapped on every plan.
Report lifecycle simplified.
reports.submitop removed. Signing and regulator submission are explicit customer-side workflow concerns — DocuSign, ServiceNow, internal GRC — not ComplyHat ops.reports.statuscolumn dropped. A report is finalized whenapproved_contentis non-null; no separate state machine to reconcile.reports.start_draft+reports.finalizeremain the only two write paths.
Phase G closed end-to-end. Hosts can now actually invoke every tool.Critical OAuth fix. Every authenticated MCP op was returning
403 INSUFFICIENT_SCOPE since the May 8 wedge. Cause: the wedge trimmed advertised scopes to ["openid", "email"] but left the server-side scope check intact. Hosts could connect and list tools, but actual op calls 403’d. Fix treats OIDC-only tokens as full access; authorization is enforced at the organization-id layer in every handler, same boundary the dashboard uses.Auditor agent host-discoverability. The guidance MCP tool now serves all six skills and six auditor agents. The fix took 5 commits across findMonorepoRoot(), build-time codegen of resources into the JS bundle, biome ignore for generated files, and a .vercelignore correction (it was stripping skills/ and every *.md before the build saw them).Stripe live. Webhook endpoint registered with all 12 events. STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, and three price IDs pushed to Vercel production. Signed-curl smoke verified the full chain: signature verify, env match, handler dispatch, idempotency insert, 200 response.Operational hygiene. Next.js bumped 16.2.3 → 16.2.6 for the May 2026 security release (moderate XSS in App-Router CSP-nonce code paths). Branch protection set on main (no force push, linear history). 'use cache' marketing pilot reclassified to WONTFIX after the auth/cache incompatibility surfaced in Next.js issue #89754. Premature prune-stripe-events cron unscheduled (free tier covers 30 years at current write rates without it). Two orphan edge functions deleted.Methodology whitepaper. Full Appendix A (8 sections, 18 citations) now live at /methodology , bias / drift / explainability / adversarial methods with thresholds tied to legal and academic sources, plus data governance, threshold rationale, and reproducibility guarantees.Six first-class agent hosts.
- Claude.ai removed from the prime-host roster. Install paths are now Claude Code, Claude Desktop, Codex Desktop, Codex CLI, OpenClaw, NemoClaw.
- Mintlify docs simplified: single-URL install per host, canonical tab order, no verbose preamble.
- DESIGN.md aesthetic locked into the docs theme (Cornflower Ocean primary, Inter, code-block themes, structured footer).
- Email infrastructure removed. Customer alerts surface via MCP
audit_events.listpolling; founder notifications via Vercel runtime logs.
Eight first-class agent hosts. Zero internal LLM calls.
- Codex Desktop and Codex CLI promoted to first-class hosts.
- Claude Desktop joins as a parallel runtime to Claude Code.
- ComplyHat is available as a remote MCP at
https://complyhat.ai/api/mcp. Paste the URL into your host config and complete OAuth consent. - AI Gateway plumbing removed entirely (migration 084). ComplyHat now makes zero internal LLM calls; host agents bring their own reasoning, ComplyHat returns structured citations only.
Generated reports clearly labeled as drafts.
- Six first-class hosts: Claude Code, Claude Desktop, Codex Desktop, Codex CLI, OpenClaw, NemoClaw.
- Human review and sign-off live in your existing compliance workflow, not in ComplyHat.
Hardening.
- IP rate limiting on signup and invite endpoints.
Compliance engines, v2.
- Explainability engine: LIME (with intercept) plus coalition-attribution proxy with narrative generation.
- Drift engine: PSI, KS test, chi-squared (three distribution tests, picked per feature type).
- Adversarial robustness and data-governance audits.
- AI Gateway only. Failover moved to Vercel AI Gateway dashboard.
Framework-agnostic pivot.
- First-class support for NemoClaw, OpenClaw, Claude Code.
- Shared
SKILL.mdandschedule.jsonformat across all 8 auditor agents.