Go-To-Market
Map any market from an ICP prompt
Describe the companies you're looking for in plain language: vertical, size, geography, funding stage. A discovery agent finds the companies that fit, treating every ICP criterion as a hard constraint and honoring a CRM exclusion list. An enrichment agent fills each company with funding, headcount, tech stack, named contacts, and buying signals. Every field carries claim-level citations and a confidence grade, and one click expands the map without repeating companies.
Quick Start
Inputs
- ICP description A plain-language profile of the companies to find, entered at run time (for example, "AI-powered vertical SaaS companies in healthcare, 11-200 employees, US or Israel, Series A or later").
- Exclusion list (optional) A CSV or text file with one domain per line, companies already in the CRM. The Mapper never returns them.
Outputs
- A Supabase table of ICP-matched companies, each with a fit reason
- 12 enrichment fields per company, from funding stage to buying signals
- A confidence grade and citation links on every field
- One-click map expansion with no repeated companies
- Natural language chat over the complete map
- CSV export and direct SQL access to all data
How it works
A 6-phase pipeline.
- Create agents A one-time script clones two gallery templates into the workspace and applies a custom output schema, goals, sources, and effort tiers via the agents API.
- Map The Mapper agent researches the ICP as an async run, minutes of real multi-source research across LinkedIn, Crunchbase, Wellfound, TechCrunch, and startup directories, and returns schema-conforming company rows, each with a fit reason and the URL where it was discovered.
- Filter and land Rows are deduplicated by domain, checked against the exclusion list and size constraints, and upserted to Supabase with the verbatim raw response.
- Enrich One enrichment run per company, fanned out with a concurrency cap. Every field arrives with claim-level citations; array fields carry one cited claim per element, so a single company can return 70 or more receipts.
- Trust Per-field grades aggregate the worst claim per field with a count breakdown. Low-confidence rows are flagged, never hidden: the dataset shows exactly what the agent was sure about and why.
- Expand and chat Expanding the map re-runs the Mapper with previous_interaction_id, same ICP context, no repeated domains. The chat tab answers questions over the full map with Claude.
Stack
Nimble primitives plus the full runtime stack.
Nimble APIs
What it does
- Web Search Agents Two custom agents built from gallery templates: one discovers companies matching the ICP, one enriches each company with cited data. Multi-turn runs power map expansion.
- Trust framework Claim-level citations and confidence grades returned with every result and stored per field in Supabase.
3rd Party Tools
Role
- langchain-anthropic Prompt template and Claude binding for the chat tab.
- claude-sonnet-4-6 Answers natural language questions over the mapped dataset.
- supabase Postgres storage for runs and companies, including per-field trust claims and verbatim raw responses. Free tier.
- streamlit UI: ICP input, live run progress, confidence badges, per-field trust panel, map expansion, CSV export.
- python 3.9+ Agent setup, run lifecycle client, and the map-enrich pipeline, including a headless CLI.
Reach out if you have any questions.
Talk to an Expert