Free today. More on the way.

Yes. The product is free under BSL 1.1 (converts to Apache 2.0 after 3 years) for any team or individual at a company under $10M ARR. You can run it three ways and mix freely: local Ollama models (no internet, no keys), your own provider keys (Anthropic, OpenAI, Google, xAI, OpenRouter — pay providers direct, no Metriq markup), or Metriq Cloud pay-as-you-go credits (top up by card, pay per token at near-wholesale). No subscription, no quotas, no upsell wall.

If you don't want to manage your own provider keys, you can fund a Metriq Cloud balance and pay per token. We route your requests through our gateway across 14+ frontier models (Claude, GPT-5, Gemini, Kimi, Qwen, GLM, MiniMax, DeepSeek, etc.) and bill at provider list price + a small handling fee that covers the gateway, billing, and trial credit. Top up by card, no subscription.

Pulse is the always-on autonomous version of the Metriq agent. Instead of you opening a notebook and asking a question, Pulse runs continuously against your connected data on a schedule you set — daily, hourly, on a metric trigger — and surfaces what it finds (revenue anomalies, churn signals, growth cohorts, weekly business reviews) into Slack and email. Custom agent goals are first-class: "watch for churn signals and alert me", "find top growth cohorts each Monday." It's a sibling product to the interactive agent, not a subscription on top of it. Pricing is custom, scaled to compute hours and the size of the data footprint.

Yes. BYOK works on every tier — we never deprecate it. The Enterprise per-seat fee pays for the platform (managed cloud, shared memory, shared skills, governance, audit, SSO/SCIM, VPC), not for inference. If you've already negotiated an Anthropic Console plan with provisioned throughput, a Bedrock contract, or a Vertex commitment, point Metriq at those keys and skip Metriq Cloud entirely. Mixed mode is fine too — Metriq Cloud for the long tail, your own keys for top providers.

The team-collaboration story Hex and Deepnote don't tell. Shared memory means a data dictionary, business rules, and query patterns the whole team agent can read — not 12 isolated copies in 12 laptops. Shared skill library means one approved monthly-revenue-cohort skill, not 12 ad-hoc reinventions. Shared connections mean credentials live once, governed by the security team, with the agent inheriting a policy-scoped view (read-only, table-allowlist, etc.). Plus the standard procurement checklist: SSO, SCIM, RBAC, audit logs, VPC deploy, SOC 2, HIPAA BAA, 99.9% SLA, indemnification.

Every action the agent takes — every SQL query against your warehouse, every file write, every Slack post, every API call to a connector — is logged at the tool-call grain with the user, prompt, model, timestamp, and result. Exportable as JSON or CSV, queryable via API, retainable per your policy. None of the BI-style notebook tools have this because they don't have an agent. It's table stakes for putting an autonomous data agent inside a regulated org.

If you BYOK with your own OpenRouter key, you pay OpenRouter directly — they charge a small fee on credit deposits, not on per-request inference, so the per-token rate is the underlying provider's list price. Metriq adds nothing on top.

If you pay-as-you-go via Metriq Cloud, our gateway routes through a mix of OpenRouter and direct provider contracts depending on the model. We negotiate volume rates with top providers and absorb the OR top-up fee on long-tail models. The handling fee disclosed at checkout covers gateway, billing, and trial credit — typically a few percent above wholesale, comparable to or below industry norms.

If you bring your own Anthropic / OpenAI / Google key directly, none of the above applies — your call goes provider-direct with zero markup.

Yes. Enterprise ships a single-tenant deploy package — Helm chart for Kubernetes, Docker Compose for simpler topologies. Your VPC, your data, your model endpoints. SSO bridges to your IdP, audit logs ship to your SIEM, and the agent never makes outbound calls except to providers you explicitly allowlist. Common pattern: deploy into an air-gapped subnet with Bedrock or Vertex as the only model endpoint.

BSL 1.1 doesn't grant production use above the threshold. In practice, teams above $10M ARR are usually moving to Enterprise anyway — they want governance, shared memory, audit trail, and SSO/SCIM, not just a single-player notebook. If you cross the line and you're still on Free, talk to us about a commercial Notebook license or upgrade. Either path lands quickly; the license is paperwork, not engineering.