Spec
The ideation agent: a research-backed thinking partner that turns ideas into buildable briefs.
You start by talking, not coding. Spec is the agent on the other side of every chat: not an intake form, not a yes-machine, but a sharp cofounder with its own research desk. It pushes on audience, wedge, and scope; verifies its claims against the live web while you talk; reads chains directly; and quietly assembles the brief your whole build starts from. By the time you hit Create project, the thinking is already done.
What it pushes on
- Specific questions: who is this for, why now, why this shape.
- Competitors and comparable products, with direct takes on weaknesses.
- Scope cuts: what belongs in v1, what waits.
- For onchain ideas: chain choice, whether a token is warranted at all, and contract surface.
It deliberately does not write code or implementation plans. That happens after the chat becomes a project.
Research, three speeds
Most chat products stop at a search box. Spec runs a research desk: three ways to check the world, picked by the weight of the question.
| Mode | When it runs | What you see |
|---|---|---|
| Quick lookups | A single claim worth checking | It searches the web and reads the actual pages mid-reply: real competitor products and official docs rather than tutorials, each step labeled with why it's looking |
| Deep research | One or two independent checks whose answers belong in this reply | The turn takes a little longer, then the findings come back woven into the same answer |
| Background research | Heavy digging: several independent questions, or a dissection that takes minutes | Queued instantly; you keep talking while researchers run behind the conversation, progress shows in the activity panel, and the findings drop in at the start of a later turn |
Research is metered in credits, and background runs announce themselves when they finish, so you're never staring at a spinner.
Onchain, read-only
For onchain questions it reads live chain data:
- Paste a contract address and it pulls the contract's profile from Etherscan: verified or not, which functions and events it exposes.
- Name a token and it fetches live market data from Mobula: price, liquidity, market cap.
These lookups are read-only by construction (the partner has no way to send a transaction), and each one costs credits, so it uses them point-blank on the address or token that matters rather than as a broad scan.
It plans while you talk
Every decision you settle ("chain: Base", "custody: agent wallet"), every question still open, and how mature each part of the idea is: the partner records all of it silently as you talk. Two surfaces read from exactly those notes:
- The maturity panel beside the chat grades concept, audience, scope, differentiator, technical, and revenue as solid, partial, or vague, so you can see at a glance what's settled and what still needs a conversation.
- The brief: when you hit Create project, the whole conversation distills into the brief that seeds the PRD. A server-side backstop records each turn even when the partner skips a note, so decisions don't get lost.
It knows you and your projects
- If a new idea overlaps something you've already built, it can list your projects and read the relevant one's summary to ground its advice ("this is close to the flow you already have in X"). It can read a project's brief, stack, build status, its written PRD, deployments, and token/contract state. It has no path to your code, environment variables, or keys, and anything resembling a secret is dropped from its notes rather than stored.
- It learns durable things about you: experience level, preferred chains, how lean you like to scope. That profile carries into future chats, so it picks up where you left off instead of re-asking the basics.
After the project exists
The same agent flips into navigator mode on a promoted chat: ask it where to do something ("where do I change the hero?", "how do I ship this?") and it points you to the right chapter, usually with a concrete prompt to use there. It advises; it never edits the project itself.
See Chats: from idea to project for the surface around it: starting, starring, attaching files, importing repos, and creating the project.