Thesis kickoff
An agent (running on Claude) checks its memory store for any prior evaluation on the name, scores portfolio overlap against the fund's IC mandate, pulls live financials via the FMP MCP server or the web, and returns a fund-shaped kickoff memo with citations. Each memo lands on one of three calls: PASS (skip), WATCH (monitor for catalysts), or DEEP DIVE (worth the next 10–20 hours).
Future state: the rendered memo is the demo's end-state, but the production end-state is a one-click "add to research tracker" that pushes the structured kickoff into the team's existing system (Notion / Sheets / Linear), seeds a recurring re-kickoff on cadence as the thesis evolves, and threads the same memory back into deal triage and holdings monitoring — so the agent's prior take on a name shows up wherever the firm next touches it. More on triage & anomaly →
Running kickoff
View current IC mandate — what the agent scores against
Stocks
- Geographies allowed: Singapore, Malaysia, Indonesia, Vietnam, Thailand, Philippines, Japan, Hong Kong, United States
- Excluded sectors: Tobacco, Defense (offensive weapons), Coal-fired power
- Min market cap tier: Mid
- Position size: $500,000 – $5,000,000
- Concentration cap per position: 10%
Privates
- Geographies allowed: Singapore, Vietnam, Indonesia, Philippines, Thailand, Malaysia, Global
- Excluded sectors: Crypto-native, Cannabis, Gambling
- Preferred stages: Series A, Series B, Growth
- Check size: $500,000 – $5,000,000
- Ownership target: 5% – 30%
View memo template — the agent's output shape
# Fund kickoff memo template
This is the template the thesis-kickoff agent uses to structure its output.
Edit it to change the agent's voice, ordering, and section depth — re-provision
the agent to pick up changes.
The template is *part of the operational layer*. A generic Bloomberg AI / Perplexity
prompt produces a generic 1-pager. This template makes every kickoff memo
identical in structure, voice, and section depth — because every memo is the
same shape, downstream review (IC, partner feedback, second-pass diligence)
gets faster the more times the loop runs.
---
## Voice
- **First-person plural** ("we") for the fund's perspective.
- **Direct, declarative**. No hedging unless the evidence supports it.
- **Quantified claims**. "Growing fast" → "growing 22% YoY (Q4 transcript)".
- **Citations on every factual claim**. Inline markdown links.
## Section structure
```
# <Ticker> kickoff — <Company name>
*<one-line description>*
## Prior evaluation
(skip header if no prior eval)
## Fit summary
2-3 sentences. Overlap with existing positions, complementary or in tension,
what owning this name does to portfolio shape.
## Business
1 paragraph. Segments, geos, where revenue actually comes from.
## Financial snapshot
Tight bullets or compact table. Revenue + growth (3y), margin trajectory,
balance sheet, capital allocation.
## Market view
Consensus EPS / revenue, recent revisions, valuation vs sector and own history.
## Bull case
3 bullets. Each: claim — evidence — citation.
## Bear case
3 bullets. Each: claim — evidence — citation. Generic bears forbidden.
## Catalysts (6-12 months)
2-4 bullets. Date / event / what it would prove.
## Risk flags
2-4 bullets. What would break the thesis.
## Fit against IC criteria
Table. criterion | score (pass/partial/fail) | one-line detail.
## Recommendation
**PASS** / **WATCH** / **DEEP DIVE** + one paragraph reasoning.
## Suggested next 3 steps
DEEP DIVE: 3 specific things to dig into.
WATCH: 1-2 triggers to revisit.
PASS: 1 sentence so this doesn't get re-pitched in 3 months.
```
## Hard rules
1. **Citations are mandatory** for every factual claim. No citation, no claim.
2. **Prior evaluation always first** — every memo opens with what we already know.
3. **Fit summary always second** — surfaces the operational differentiator.
4. **Memo fits in ~2 pages** rendered. If it doesn't, the agent is padding.
5. **Bear case is the most important section**. A weak bear case breaks trust
with the partner. Match the bull case in specificity.