17 pages, 32 API routes, real-time WebSocket risk feed, Monaco-powered policy editor with hot reload, custom rules DSL, agent registry with scoped credentials, kill switch, audit trail with PII tokenisation, RBAC, OIDC SSO. Hosted at app.clampd.dev or self-hosted in your VPC.
Dashboard is the operational surface: kill an agent right now, suspend one, edit a policy, approve a tool descriptor, change a threshold, manage members. Real-time, action-oriented, used during incidents and during normal ops by SecOps and SREs.
Two adjacent surfaces are being split into separate products. Clampd is source-available under BSL-1.1 with a Free tier (Design Partner) and an Enterprise tier; the raw audit data lives in your own ClickHouse / Postgres on every tier and is queryable directly. The maintained workflows on top are Enterprise.
You can rebuild any of the Enterprise workflows yourself from the raw data — most teams find the maintenance overhead isn't worth it, which is what the paid tier is for. There's no encryption-with-vendor-key gating; your data is yours either way.
Every operator action goes through ag-control. The dashboard is a thin UI on top of an audited control plane.
/Live operational view. Active agents, recent denials, top matched rules, current EMA risk distribution, throughput.
/agentsAgent registry with scoped credentials. Per-agent state (active / suspended / killed), boundary config, recent calls.
/killEmergency kill switch. Per-agent kill, descendant cascade option, per-layer cascade results, idempotent.
/policiesCedar policy editor with Monaco, syntax highlighting, validation, hot-reload across the fleet within seconds.
/rulesCustom rules DSL. Browse the 263 built-ins by category, OWASP LLM tag, or regulation. Add / disable / score-tune.
/tool-descriptorsDiscovered tools awaiting category assignment. SHA-256 descriptor hashes for rug-pull detection. Approve/deny.
/audit-trailSQL-queryable ClickHouse audit. Filter by tool, agent, denial reason, time. PII automatically tokenised.
/delegationAgent-to-agent delegation graph. Approved peer relationships, depth limits, cycle detection.
/delegation-analyticsPer-pair call counts, cross-agent risk patterns, escalation chains over time.
/thresholdsPer-org tuning of risk thresholds (block, flag, auto-suspend EMA), gray-zone bounds for LLM judge.
/scope-exemptions13 exemptable rules, 62 never-exemptable. Granted scopes per agent, expiry, full audit on every approve.
/workflowsApproval workflows for high-risk actions. Slack approval, multi-step gates, expiry.
/protected-entitiesTag specific resources (DBs, repos, customers) as protected; policies fire when an agent touches them.
/keywordsCustom keyword dictionaries — your industry's PII patterns, internal product names, competitor mentions.
/api-keysOrg-scoped API keys with permissions (read / write / kill). Rotation, last-used timestamps.
/bundlesPre-built rule packs (industry-specific, OWASP-aligned, compliance-driven). Install with one click.
/settingsOrg config, OIDC SSO (Okta / Azure AD / Keycloak), members + RBAC, billing, webhooks, Slack/PagerDuty.
Modern stack with strict separation of concerns. The dashboard is the operator-facing surface only; it does not directly write to runtime state.
The dashboard never writes directly to Redis. Every runtime state mutation (kill, suspend, scope grant, policy update) goes through ag-control over WebSocket / HTTP, which validates and emits the audit event. This means a compromised dashboard can't bypass the audit trail or skip cascade layers — every operator action is reproducible from logs.
Both options run identical code. The hosted version (app.clampd.dev) is appropriate for teams that want zero infrastructure. The self-hosted version is for regulated industries, air-gapped environments, or teams that already have Kubernetes / Docker Compose practice. Stack, schema, and feature parity are the same.
Hosted at app.clampd.dev — JWT auth, free tier under 25 agents, no credit card. Self-hosted via Docker Compose with full feature parity.
Sign In to Hosted → Self-Host