Sightspool builds a persistent, source-backed understanding of your product and turns it into tracked assumptions, cited findings and client-owned actions. This page takes you from an empty workspace to your first citable read.
1. Create a workspace
Sign up at sightspool.com — every new workspace starts on Builder Free. You get sample data to explore, a small monthly AI allowance, and no clock: exploring sample data does not start a trial, and we do not collect a card.
2. Connect a source
On the Connections page, connect the sources your evidence should come from. PostHog is first-class: Sightspool reads your project through your own API key, encrypted at rest and scoped to your workspace. A GitHub repository connection lets the register react to what you ship; Stripe adds the commercial context behind behavioural reads.
3. Run the first analysis
Hit Run analysis on the Connections page. The Sightspool Agent reads your connected sources and posts what it finds to the Evidence feed. Every metric or alert it publishes must cite the live tool call that produced the number — findings without proof are structurally rejected, not merely discouraged. That is the product's first rule: no proof, no post.
4. Work the register
The Assumptions page is the register: each entry is a belief made falsifiable — a question, a signal definition, a threshold, and a verdict with append-only history. The agent drafts entries from your repo and your evidence; committing one is always your call. Committed entries go under watch: their signals run against your connected sources on a schedule, and below the minimum sample the register says "still gathering" — it never upgrades a thin read to a verdict.
5. Talk to the agent
The agent is ambient: open the chat from any page ("Ask the agent"), and the thing you're looking at — a register entry, a finding — pins itself as context. "Go deep" runs a full studio session that routes your question through research, interaction-design and service-design lenses and synthesises an answer. When the lenses disagree and no data settles it, the session escalates to a human instead of bluffing.
Where things live
- Overview — the signed-in home: what changed, what's waiting on you.
- Feed — proof-gated findings, with the cited evidence one click away.
- Assumptions — the register: beliefs, thresholds, verdicts, watch state.
- Interventions — actions proposed by the agent that only you can approve.
- Connections — sources, the in-product SDK key, and coding-agent (MCP) access.