Sigil
Handshake deals into binding micro-contracts
An AI-native mobile app that turns a spoken agreement into a binding digital micro-contract, with the language model running on the device so the conversation never leaves the phone.
- Where
- Sigil — founder
- When
- 2026 — present
- Role
- Founder — product, architecture, roadmap
- Stack
- On-device SLMs · Speech processing · Mobile · Privacy-first architecture · Subscription product
Problem
Most freelance work still starts as a handshake — a voice note, a call, a verbal yes. None of it is enforceable, and the tooling that would make it enforceable is slow, expensive and built for enterprise legal teams rather than for one person who needs cover before starting a job tomorrow.
Constraints
- 01Voice data cannot leave the device — this is a privacy product or it is nothing
- 02Zero-latency processing, because dictating a deal should feel like talking
- 03A consumer product with a subscription, deliberately not a custom agency service
- 04Bootstrapped: every architectural choice has to be affordable to run at zero revenue
Architecture
Key decisions & trade-offs
A small model on the device, not a large one in the cloud
A cloud LLM would be more capable per token. But it would also mean uploading a recording of a private negotiation, adding a network round-trip to something that should feel instant, and paying per inference for a product with no revenue yet. An SLM on-device answers all three at once.
Trade-offYou give up raw capability and you inherit the device's memory and thermal limits. The scope is narrow enough — extract terms from a short spoken agreement — that a small model can carry it.
A consumer product, not an agency
The fastest revenue for a solo founder with these skills is bespoke client work. That is also a business that cannot compound. Sigil is deliberately a subscription product so the work done this month still earns next month.
Trade-offMuch slower to first revenue, and it means saying no to consulting that would pay immediately. It is the difference between a job and a company.
Human confirmation before anything binds
The model drafts; it never commits. Both parties see the extracted terms in plain language and confirm before a contract exists. An extraction error should cost a correction, not a legal dispute.
Trade-offIt adds a step to a flow whose entire pitch is speed. Given the output is legally binding, that step is not optional.