discover-investors

name	discover-investors
description	Discover a batch of new investor targets from the public web and propose them into the round.sh CRM. Use when the founder wants to build or grow the pipeline — "find pre-seed B2B funds in NYC", "build a target list of 30 angels in fintech", "fill out the pipeline with seed AI infra investors". Outputs round CLI proposals, never writes the CRM directly.

Discover Investors

Bulk discovery + enrichment of investor targets via Exa Websets, landing each as a round investors add proposal the founder approves. This is the pipeline-building counterpart to the single-firm investor-researcher agent.

When to use (and when not)

Use this when the founder wants new rows in the pipeline from a profile, e.g.:

• "Find seed-stage B2B SaaS funds in NYC writing $250k–$1M checks."
• "Build a target list of 30 fintech angels."
• "Who else should be in my pipeline for an agent-infrastructure seed round?"

Reach for a sibling instead when:

• They want to deep-research one firm/partner already in mind → the investor-researcher agent (via /prep or directly).
• They want generic company/people research, not investors-into-the-CRM → company-research / people-research.
• They want a CSV of sales leads (customers to sell to, not investors) → exa-lead-gen.

Tools

• exa-websets MCP (ships in this workspace's .mcp.json) — async discover → enrich → verify loop. Load via ToolSearch if absent: mcp__exa-websets__preview_webset, mcp__exa-websets__create_webset, mcp__exa-websets__get_webset, mcp__exa-websets__list_webset_items.
• The round CLI (via Bash) to propose rows. Proposal-only: never --yes, never write round.db directly.

Flow

1. Pin the ICP, then confirm before spending API. Settle on entity (funds/CVCs → company; angels → person), stage, sector, geography, check size, and target count. Echo it back and get a yes — Websets calls cost money.
2. Preview the parse. Call preview_webset with the natural-language query (e.g. "Pre-seed B2B SaaS funds in NYC writing $250k–$1M checks") and show the founder the parsed entity + criteria. Adjust if Exa misread it.
3. Create the webset. create_webset with searchQuery (confirmed), searchEntity, searchCount (default 25; confirm if > 50), searchCriteria (what the founder emphasized), and enrichments — one per field you want filled, mapped to a settable CRM column (see the table below).
4. Wait for idle — in a spawned Task agent. Polling get_webset until status: "idle" and reading items is token-heavy; do it in a Task subagent that returns only the distilled rows, so the main context stays lean.
5. Dedupe against the existing pipeline. Pull round investors list and match by name (case-insensitive; strip "Capital"/"Ventures"/"Fund"/"LP"). For a match, enrich the existing row with round investors update --id <n> ... rather than proposing a duplicate.
6. Show a preview table, then propose. One round investors add per genuinely new target.

Mapping Websets output → round investors add

Only these flags are settable from the CLI — map enrichments to them:

round investors add --name "<firm>" \
  --type "Fund|Accelerator|Angel|Strategic / CVC" \
  --tier <A|B|C> --fit <1-10> \
  --hq "City, Country" --check-size "$250k–$1M" \
  --website <url> --linkedin <url> --x <handle> \
  --fund-vintage "Fund III, $80M, 2024" \
  --talking-points "<thesis fit in their language, 1-2 lines>"

Richer fields the CLI has no flag for — stage, why-fit rationale, notable bets, warm path, lead capability — go into the investor's investors/<slug>.md profile or are edited in-app after the row lands. Note them in the preview; don't pretend a flag exists.

Honesty & guardrails

• Proposal-only. Every round investors add records a proposal in the approval queue. Never pass --yes unless the founder explicitly says apply now. Never touch round.db directly.
• Don't fabricate fit. Score 1–10 only from real signal; if enrichments don't support a score, leave it off and say so. A low/honest score is useful.
• Don't propose dupes or overwrite founder-curated rows (status, tier, warm-path, notes) on enrich-updates.
• Don't push contacts you can't verify publicly. Prefer a public [e-mail] or leave --contact empty.
• Keep the founder in the loop: discovery proposes; the founder approves in the app.