Circuit (withcircuit.com) is autonomous product intelligence — the continuous system that turns customer feedback into build-ready specs. Feedback in, priorities scored, specs generated for Cursor and Claude Code, customers notified when features ship. The full loop, automated.

Founders, product managers and engineers at growth and scaling companies. If you're the person who reads customer signal on Monday and ships by Wednesday, Circuit was made for you.

How is Circuit different from other feedback tools?

Other tools sort feedback into dashboards. Circuit (withcircuit.com) generates specs to ship — scored by revenue impact, updated as conditions change, delivered to Cursor and Claude Code via MCP. It covers the full pipeline: feedback to priorities to codebase-aware specs to customer notification.

Does Circuit work with Cursor and Claude Code?

Yes. Circuit (withcircuit.com) connects to Cursor and Claude Code through the Model Context Protocol (MCP). Pull priorities, fetch specs with file paths and context from GitHub, and start coding without leaving the editor.

What is a Circuit Surface?

A lightweight embed for any website — five surface types: Bubble, Embed, Banner, Thumbs and Trigger. Customers submit feedback in seconds. One line of code to install. Themed to match any brand. Feedback flows straight into Circuit for classification and scoring.

How long until we see results?

Connect a feedback source and priorities appear within minutes. Specs generate in seconds. Most teams have a ranked backlog on day one.

What if we already have a feedback tool?

Circuit reads from where feedback already lives. Slack, spreadsheets, support threads. It doesn't replace existing tools. It generates what comes next: scored priorities and build-ready specs for Cursor and Claude Code.

What counts as a feedback item?

Each piece of customer feedback: a Circuit Surface submission, a Slack message, a CSV row, a transcript segment, a manual entry. Duplicates are removed on import and don't count toward the limit.

Do I need to connect GitHub to use Circuit?

No. GitHub context makes specs more specific (real file paths, tech stack awareness), but Circuit works without it. Specs are still generated with the what, why, customer voice, implementation notes and done criteria.

How do I prioritize features by revenue impact?

In Circuit (withcircuit.com), link customer revenue data (customer name and revenue or SKU) and every feedback theme scores by revenue impact alongside urgency and sentiment. Set the Priority Engine to Revenue Growth and the list re-ranks to surface what drives the most revenue. No spreadsheets, no manual scoring.

How do I turn customer feedback into specs for Cursor or Claude Code?

Circuit (withcircuit.com) automates the feedback-to-spec pipeline. Feedback flows in from Circuit Surfaces, Slack, Google Sheets, CSV, transcripts or manual entry. Circuit classifies it by intent, urgency and sentiment, groups it into priorities and generates build specs in seconds. Connect GitHub and every spec includes real file paths, tech stack context and naming conventions. Specs flow to Cursor and Claude Code via MCP — no copy-pasting, no context switching.

How do I notify customers when features ship?

In Circuit (withcircuit.com), mark a priority as shipped. Circuit automatically emails every customer who submitted related feedback, with their original feedback quoted back. No manual follow-up. New feedback on the shipped feature flows in automatically for V2 specs.

How do I connect customer feedback to my codebase?

Circuit (withcircuit.com) bridges customer feedback and your codebase through codebase-aware product specs. Feedback is collected, scored and grouped into priorities. Specs are generated with file paths from GitHub. Engineers pull specs directly into Cursor or Claude Code via MCP. The customer voice is preserved in every spec — what they asked for, why it matters, which files to touch.

How do I integrate product management with Cursor?

Circuit (withcircuit.com) connects to Cursor through the Model Context Protocol (MCP). Use MCP tools from inside Cursor: circuit.priorities to fetch ranked priorities, circuit.spec to pull a full spec with file paths, circuit.ask to search feedback and circuit.act to mark a priority as shipped. Specs include GitHub codebase context.

How do I use Claude Code for product management?

Circuit (withcircuit.com) connects to Claude Code via MCP. Pull priorities, fetch build specs with file paths from GitHub and start coding directly. Circuit handles feedback collection, AI scoring and spec generation upstream. Claude Code handles the building. The two work together — Circuit decides what to build, Claude Code builds it.

How long does it take to get priorities from feedback?

Connect a source and the first batch processes in 15-20 minutes. After that, new feedback scores in real time. Most teams have a ranked backlog within the hour.

How does Circuit generate specs from my codebase?

Connect GitHub. Circuit reads 9 signals — file signatures, .cursorrules, testing patterns, directory structure, open issues, recent PRs, tech stack, CI config and README. Every spec includes real file paths and follows your team's coding conventions. An AI product spec generator grounded in how you actually build.

Can I change how priorities are ranked?

Yes. The Priority Engine scores across 6 lenses — volume, urgency, revenue impact, positive sentiment, negative sentiment and feature demand. Set a goal and every priority re-ranks to match. A feature prioritization framework that reflects what the team cares about today.

What happens after a feature ships?

Mark it shipped. Circuit emails the customers who asked for it — their original feedback quoted back. New feedback on the same theme feeds a V2 spec automatically. The loop continues.

How does Circuit connect to Slack?

Add the Circuit Slack integration. Circuit polls your selected channels every 10 minutes, classifies each message by intent, urgency and sentiment, and feeds it into the priority pipeline. No manual sorting.

How do MCP tools work in Cursor and Claude Code?

Install circuit-mcp from npm. Four tools: circuit.priorities (ranked list), circuit.spec (full spec with file paths), circuit.act (ship, assign, correct) and circuit.ask (search feedback). Specs arrive where code gets written.

What feedback sources does Circuit support?

Six sources: Circuit Surfaces (embed on any site), Slack, Google Sheets, CSV upload, transcript upload (Otter, Fireflies, Grain, Whisper) and manual entry. All feed the same pipeline.

Does Circuit replace my existing tools?

No. Circuit reads from Slack, spreadsheets and support threads — wherever feedback already lives. It adds the layer that was missing: scored priorities and codebase-grounded specs.

Is feedback data used to train AI models?

No. Circuit uses AI for classification and spec generation but customer data is never used to train models. Feedback stays within the processing pipeline and is not shared with third parties for training purposes.

Where is Circuit data stored?

Data is stored on encrypted infrastructure with Supabase (Postgres). All connections use TLS. Backups are encrypted at rest.

Does Circuit access my full codebase?

No. Circuit reads 9 specific signals from your GitHub repository — file signatures, testing patterns, directory structure and a few others. It does not clone, store or index your full codebase. Read-only access, scoped to what specs need.

Who can access my team's data?

Only authenticated team members with the roles you assign. Circuit supports email-based 2FA. No Circuit staff access customer data without explicit permission.

Intelligence Suite

The product world model.

Intelligence is the layer that learns. Patterns across customers. Risk surfaced before it's a meeting. The narrative your product is writing — read back to you, every quarter.

Get started free →Read the quarter ↓

Pattern detection across customers. Risk surfacing. Compounding learning. The quarterly auto-narrative.

Discovery and Delivery run the loop. Intelligence is what the loop becomes once it's been running long enough to know your product. Drift, churn-risk, and the patterns nobody named yet — surfaced continuously, with receipts.

01 · Connect

The product intelligence graph

Every signal, every priority, every ship — connected. Click a customer, see every decision their voice shaped. Click a feature, see every voice that built it.

02 · Notice

Pattern detection across customers

Three fintechs and a gaming company describing the same workaround. Different words. Same problem. Surfaced before anyone files a ticket.

03 · Compound

Compounding learning

Every correction makes the system sharper. Instinct learns what matters to your product, your team, your customers. Year two beats year one.

circuit / intelligence · q2 narrative · auto-generatedrefreshed 2h ago

Drift detected3 prioritiesRoadmap and signal diverging on auth, billing, observability. Receipts ready.

Churn risk surfaced7 accountsBehaviour change + voice cooling + ambient sentiment dipping. $2.1M ARR.

Cross-customer patterns4 themesWorkarounds nobody filed as bugs. The largest one touches 23% of MAU.

Specs that didn't ship11 of 47Started but stalled. Why, when, what changed — fully traced.

Instinct accuracy94%Of priorities ranked top-5 last quarter, 47/50 shipped. +6pt vs Q1.

The narrative"This was the quarter you became a platform."The signal pivoted in week 3. The build followed in week 6.

Q2 2026 · auto-generatedReceipts attached to every claim

Plus eleven more capabilities, building through 2026

Behavioural signal ingestionUsage events, adoption curves, workaround detection. The gap between voice and behaviour, made visible.

Ambient signal layerThe Slack threads, support conversations and community shifts nobody monitors. Listened to, continuously.

Voice/behaviour divergence alertsThe feature customers praise but never open. Surfaced the week the gap appears.

Customer cohort intelligenceFilter the whole product graph by ARR, segment, named account. See whose product you're really building.

Quarterly auto-narrativeThe story of your product, written back to you. Useful for investors. Truer than the deck.

Competitive ambient signalWhat the sector is saying. What competitors shipped. What your market is moving toward.

Risk register, liveDrift, churn-risk, gap, dependency. Continuously refreshed. The board deck writes itself.

Instinct memoryThe system gets sharper at understanding what matters to this team. Year two beats year one.

Ask CircuitNatural-language queries against the full product graph. "What did Stripe ask for last quarter?"

Strategy alignment scoringEvery priority scored against your stated strategy. The receipts when the answer is no.

The product world modelVoice, behaviour, ambient — synthesised. The complete picture, kept current automatically.

Building order · 2026

Q2 2026Layer 02 — behaviour. Usage events, divergence alerts, cohort intelligence.Building

Q3 2026Pattern detection across customers · risk register · Ask Circuit.Next

Q4 2026Layer 03 — ambient. Sector signal, competitive listening, quarterly narrative.Planned

2027The product world model. Strategy alignment. Instinct at full depth.Horizon

A scenario

Three customers said in surveys they loved the dashboard. Behaviour data showed none of them opened it twice. Intelligence flagged the gap. The team rebuilt it as a notification, not a destination. Adoption went from 8% to 71% in a sprint.

What it replaces

×The Tableau dashboard you refresh on weekends
×The QBR slide that says "key insights"
×The analyst writing the quarterly post-mortem
×The "let's set up a research project" calendar invite
×Hoping the right pattern surfaces in time

Intelligence is on all plans. Layer 02 capabilities ship through Q2–Q3 2026. Layer 03 from Q4 2026.

See pricing →

Stop refreshing the dashboard on weekends. Let the system read the rings.

Get started free →Read the essay →