Anthropic, the skeptical founder
Pressure-tests the idea, looks for the wedge, calls out weak demand, and asks what should be cut before anyone builds.
IdeaClyst · local-first planning instrument
Rough ideas become
buildable plans.
A structured council between OpenAI and Anthropic runs on your machine and writes a plan you can build from.
Council OpenAI + AnthropicOutput a founder packet in five partsRuns on your own disk
What goes in
A rough ideaA title and a few sentences
CompetitorsScouted before the models speak
Demand signalsComplaints, fresh launches and gaps
Source notesEvidence that grounds the council
IDEACLYSTTwo voices, one verdictideaclyst.comResearch · Council · Packet
What comes out
SummaryThe product, the customer, the reason
MVP backlogOrdered, ready for a coding agent
RisksAssumptions paired with mitigations
Validation testsExperiments, metrics, kill criteria
Next promptsHandoffs for the next week of work
Files on diskReadable Markdown artifacts
The council
Two voices, one verdict.
One prompt gives an answer. The council gives each model a job, makes them read each other, and argues until the recommendation is clear.
OpenAI, the pragmatic CTO
Designs a buildable MVP, names the data model and services, and flags any strategy that becomes expensive in code.
The argument has to resolve.The packet keeps the tradeoffs but chooses a direction and the next proof step.
The run
Research first, then five council steps.
Each step writes to disk as it completes. The run is local-first, readable, and degrades gracefully when a source is unavailable.
- 00Scout the marketCompetitors, demand signals, and source notes ground the council before the models speak.
- 01Shape the wedgeAnthropic turns the raw idea into a focused customer, offer, and risk map.
- 02Design the MVPOpenAI designs the smallest useful release and calls out technical traps.
- 03Attack the architectureThe product lens asks what is too large, too slow, or not tied to proof.
- 04Attack the strategyThe engineering lens asks which assumptions are costly or infeasible for an MVP.
- 05Deliver the packetThe debate resolves into a recommendation, backlog, risks, and validation tests.
The packet
A founder packet in five parts.
Structured for the next week of work, not for a slide deck nobody opens again.
Summary
The recommended product, the customer, and the reason to build this version first.
MVP backlog
An ordered list of the smallest valuable release, ready for a coding agent.
Risks
The assumptions that can break the plan, paired with practical mitigations.
Validation tests
Concrete experiments, success metrics, and kill criteria for the first proof loop.
Next prompts
Prompt handoffs for product, design, engineering, and follow-up research.
Files on disk
Every run writes readable Markdown artifacts locally, including the founder brief.
Research
The council starts with evidence.
Dossier
Sources become a competitor matrix, opportunity map, validation experiments, and a one-page founder brief.
Idea discovery
Name a market and a goal. IdeaClyst looks for complaints, fresh launches, and gaps, then ranks buildable candidates.
Promotion to council
Only ideas with evidence become council-worthy. The selected report carries its thesis into a full planning run.
The workspace
Find, validate, build, and learn.
IdeaClyst keeps a local library of ideas, reports, decisions, validation outcomes, and build-ready specs.
Evidence ledger
Claims, experiments, and validation outcomes move the score. Stale evidence gets sent back to research.
Decision log
Open assumptions and council disagreements become inspectable memory for the next run.
Watchtower
Monitor competitor pricing and positioning changes, then score the ones that affect your assumptions.
Build queue
A council verdict compiles into requirements, tasks, GitHub issues, and visible status.
Local-first
Local-first by design.
Runs are stored on your own disk as JSON and Markdown. IdeaClyst does not handle API keys: OpenAI and Anthropic authenticate through their own local sessions.
contact@ideaclyst.com