What is Torno?

Torno is the agent-native system of record for growth work on digital products. It does not build your product; it maintains the feedback system around it, so humans and agents can decide what growth work to do, execute the measurable parts, and learn whether any of it worked.

The Loop is the core object

Torno’s primary noun is the Loop — not a dashboard, not an event stream. A Loop is a durable state machine that connects Signals → Analyzers → Insights → Hypotheses → Changes (plus their exposure) → Results → Decisions → new Signals, with one target Metric and guardrail Metrics at the center. Every object in that chain has stable identity, provenance, and lifecycle state, and all of it is addressable by agents.

A Loop moves through listening → diagnosing → proposing → pending_approval → live → measuring → decided, then returns to listening. A Decision is one of ship, kill, iterate, or extend, recorded with rationale rather than inferred from silence. The Loop covers the full lifecycle; the Loop essay explains why the cycle, not the chart, is the unit of growth work.

Who Torno is for

Founders and growth or product teams at seed-to-Series-B digital-product companies, including solo builders — and the coding and marketing agents working on their behalf. The operating assumption is explicit: agents operate, humans govern. Capabilities are registered once and surface through the web app, the public API, the CLI, and MCP from the same contracts, so an agent is not working with a reduced subset. Irreversible, sensitive, and money-spending actions pass through policy and approvals no matter who asks.

A Property can be an app, a marketing site, or a pre-launch idea. A Workspace can hold several.

What Torno bundles

Torno bundles the instruments a Loop needs and keeps one semantic and evidence model across them. The goal is not to beat every specialist tool at its standalone job; it is that your funnel, your Experiment, and your revenue readout agree on what words mean.

The constraints Torno is built around

• Honest measurement. Every readout carries status, evidence state, bounds, data basis, and freshness. “No data”, “below the cohort minimum”, “provisional”, and “insufficient evidence” are different answers, and none of them is rendered as zero.
• One number, one definition. Every consumer of a Metric — funnels, Experiments, the Needle, the CLI — reads the same versioned definition. Nothing recomputes a named Metric on the side.
• Two-zone identity. Anonymous visitors stay cohort-only. A Subject key is join plumbing, not a profile. Identified Customers exist only after signup, in profile mode, and agents cannot read them by default.
• Untrusted content stays data. Public-web and voice content is quarantined, provenance-labeled, and fenced everywhere a model or agent could see it. It is never treated as instructions.
• Human governance. Irreversible, sensitive, or spending operations pass through policy and, where required, a human Approval Request. Agent reasoning is stored as a claim, never followed as a command.
• Critical-path safety. Assignment runs locally from signed cached bundles and fails open to control. Torno adds no network dependency to your request path.
• No lock-in. Public reads, webhooks, and workspace export are part of the model. Imports are on-ramps, not the strategy.

What Torno is deliberately not

Version 1 draws hard boundaries. Torno does not include session replay, a general customer data platform, a hosted page builder, or a full feature-flag management suite. There are no mobile SDKs yet. Payment and ad-auction rails stay with their providers. And although listening reads the public web, Torno does not automate outreach to the people it finds there.

These are product boundaries, chosen so the Loop stays the center of gravity. For finer-grained scope questions, see the FAQ.

Where to go next

• Getting started — from a repository to a first Loop.
• The Loop — states, evidence, and what closing a Loop means.