A free giveaway. This is the same system used to run investor outreach for a live capital raise: over $5M raised, and 50+ investor meetings in the first 60 days.
Follow this guide end to end and you will have it running on your own machine in about fifteen minutes. No database, no API keys, no signups.
Six agents and nine skills that run every part of a capital raise from your terminal.
The six agents are the team. You talk to them like people.
| Agent | What it does |
|---|---|
investor-researcher |
Finds family offices, institutions, HNWIs, funds, and strategics. Qualifies them against your thesis. |
investor-enricher |
Finds the decision maker at each firm, verifies contact routes, and cross-references your network for warm paths. |
outreach-strategist |
Segments investors by how strong the real connection is, assigns senders, writes the copy. |
fundraising-copywriter |
Investor memo, one pager, FAQ, follow-ups, data room. |
campaign-manager |
Builds and stages sequences, then classifies and routes every reply. |
pipeline-tracker |
Every conversation, every commitment, every overdue follow-up. |
The nine skills are what the agents actually run. You can call them directly too.
lp-discovery · thesis-matching · contact-verification · warm-path-finder
· outreach-writer · outreach-sequencer · reply-classification ·
pipeline-tracker · data-room-prep
Everything runs through conversation. Your data stays in local files on your machine.
| Thing | Cost | Required? |
|---|---|---|
| Claude Code | Claude subscription or API usage | Yes |
| A folder on your computer | Free | Yes |
| Your LinkedIn connections export | Free | Strongly recommended |
| Everything else | — | No |
That is the whole list. There is deliberately no database, no MCP server, and no API key to configure. Optional upgrades are at the bottom of this guide, for when you outgrow files.
Claude Code is Anthropic's official terminal app for Claude. It runs everything here.
Follow the official instructions at anthropic.com/claude-code. macOS, Windows, and Linux are all supported.
Then check it works:
claudeYou should get a welcome screen and a prompt. Type /login and authenticate if
you have not already. Exit with Ctrl+C twice.
Two steps: clone the repo, then copy the pieces where Claude Code looks for them.
git clone https://github.com/corebits-io/fundraising-team.git
mkdir -p ~/.claude/skills ~/.claude/agents
cp -r fundraising-team/skills/* ~/.claude/skills/
cp -r fundraising-team/agents/* ~/.claude/agents/git clone https://github.com/corebits-io/fundraising-team.git
New-Item -ItemType Directory -Force "$HOME\.claude\skills", "$HOME\.claude\agents"
Copy-Item -Recurse -Force "fundraising-team\skills\*" "$HOME\.claude\skills\"
Copy-Item -Recurse -Force "fundraising-team\agents\*" "$HOME\.claude\agents\"Do not delete the cloned folder. The skills read the references/ files from it,
and you will copy the templates/ in a moment.
Start Claude Code and run:
/skills
You should see all nine: lp-discovery, thesis-matching,
contact-verification, warm-path-finder, outreach-writer,
outreach-sequencer, reply-classification, pipeline-tracker,
data-room-prep.
Then:
/agents
You should see all six.
If either list is short, see Troubleshooting below.
This is where your actual data lives. Keep it separate from the cloned repo.
mkdir ~/my-raise
cd ~/my-raise
cp ~/fundraising-team/templates/raise-profile.md .
cp ~/fundraising-team/templates/investors.csv .
cp ~/fundraising-team/templates/pipeline.csv .Windows:
New-Item -ItemType Directory "$HOME\my-raise"
Set-Location "$HOME\my-raise"
Copy-Item "$HOME\fundraising-team\templates\raise-profile.md" .
Copy-Item "$HOME\fundraising-team\templates\investors.csv" .
Copy-Item "$HOME\fundraising-team\templates\pipeline.csv" .Never put this folder in a public repository. It will contain investor names, contact details, and notes from private conversations.
This is the most important step in the whole guide. Do not skip it and do not rush it.
raise-profile.md is the file every skill reads. It is where the system learns
what you are raising, who you are looking for, what is true about your traction,
what you are not allowed to say, and who is sending.
Open it and fill it out honestly. Ten minutes here saves hours later.
Three rules while filling it out:
- If something does not apply, write
N/A - If you do not know yet, write
TBD— the agents will ask rather than invent - Never make up a number to fill a blank
Section 1 matters most. It asks which exemption you are raising under. If you are US-based and do not know whether you are 506(b) or 506(c), stop and ask your securities counsel before going further. Under 506(b) you generally cannot cold contact strangers at all, and the system will refuse to build you a cold list. It will point you at your warm network instead, which is the correct behaviour.
If you would rather do this conversationally, just start Claude Code in the folder and say:
help me fill out my raise profile
Start Claude Code from your raise folder:
cd ~/my-raise
claudeAlways do this before anything cold. It is the highest-converting step in the entire system.
Export your LinkedIn connections first: LinkedIn → Settings & Privacy → Data
Privacy → Get a copy of your data → Connections. It arrives by email in a few
minutes. Save Connections.csv into your raise folder.
Then:
who in my network can introduce me to investors?
find investors for my raise
The researcher will confirm your brief, show you a search plan, and then work through several angles. It stops and shows you samples along the way. Aim for 150 to 400 genuinely qualified investors, not thousands.
qualify this list against my thesis
You will be shown a rubric before anything is classified. Read it. The competitor definition and the signal keyword list are the two things that most commonly need correcting, and correcting them here saves a rebuild later.
find the right contact at each firm
write outreach for these investors
You get segments, senders, and three variations each. Every message is checked against the voice rules and the compliance rules before you see it.
build me a sequence for the first 30
Nothing sends. The system stages a queue and hands it to you. You send, from your own tool or by hand. This is deliberate and it is not configurable.
here are the replies I got: [paste them]
what needs follow up today?
give me a weekly summary of the raise
You can also call an agent by name for a bigger job:
use the investor-researcher agent to build me a list of family offices
that have backed anything in climate infrastructure
use the campaign-manager agent to process this week's replies
The agents run the skills. Either route works. Talk normally and the right one gets picked up.
These are built in on purpose and they are not configurable.
Nothing sends automatically. Ever. Sequences are staged. You send.
Every irreversible step has an approval gate. You see samples and counts before anything is written.
No fabrication. Not found is the correct answer for anything unverified.
Every claim carries a source URL. Email addresses are never guessed from
patterns, because a bounce spike on a young domain can end your ability to send
email at all.
Compliance guardrails on every message. No projected returns in cold outreach. No guaranteed tax outcomes. No assuming the reader is accredited. Nothing pending described as done.
The solicitation gate. If your raise profile says 506(b), the system refuses to build you a cold list and explains why.
None of this is legal advice. Talk to securities counsel before you send anything.
Check they landed in the right place:
ls ~/.claude/skills/You should see nine folders, each containing a SKILL.md. If you see a single
fundraising-team folder instead, the copy command copied the parent rather than
the contents. Re-run it with the /* on the end.
Restart Claude Code after fixing.
ls ~/.claude/agents/You should see six .md files. Same fix as above.
The skills read references/ from the cloned repo. If you moved or deleted it,
tell Claude where it is, or re-clone.
Check section 1 of your raise profile. If it says 506(b), that refusal is
correct. Under 506(b) you generally cannot solicit strangers, and building a cold
list you cannot legally message would waste your time and put your exemption at
risk. Use warm-path-finder instead.
If it says TBD, fill it in.
That is the system working. It will not invent an email address, a check size, or an investment thesis. A blank field is recoverable. A confident wrong one gets you a bounce, a bad meeting, or a credibility problem.
Real research is slow. A 200-investor list built properly takes a few sessions, not one prompt. Speed here comes at the cost of accuracy, and an inaccurate investor list is worse than no list.
Use forward slashes or quoted paths, and tell Claude your paths explicitly. If
something looks wrong, say my raise folder is at C:\Users\me\my-raise and it
will adapt.
No. The whole system runs on Claude Code and local files. Optional paid tools are listed below, and they buy you speed, not accuracy.
It stays in local files on your machine. Nothing is uploaded anywhere except your conversation with Claude. Keep your raise folder out of any public repository.
The workflow applies anywhere. The compliance rules in references/compliance.md
are written around US exemptions, and every jurisdiction differs, often more
strictly. Talk to local counsel and adjust the guardrails.
No, and it will not be persuaded to. It stages queues, you send. An automated send loop attached to a language model and a securities offering is a bad idea everywhere.
150 to 400 for a first raise. A list of 5,000 unqualified rows is worse than 150 good ones. It costs more to message, damages your sending reputation, and buries the twelve real conversations in noise.
Drop your CSV in the raise folder and say qualify this list against my thesis.
It will map your columns and run the rubric.
Yes. Everything is a markdown file. Edit references/voice-and-copy-rules.md to
match your own voice, or references/classification-rules.md to fit your market.
The compliance guardrails are the ones to think hard about before loosening.
Yes. raise-profile.md handles both. Set the raise type in section 1 and the
skills adapt.
You will not need any of this on day one. Add it when volume actually demands it.
| Tool | What it adds | When |
|---|---|---|
| Serper / Exa | Programmatic search at volume | Past ~500 targets |
| Hunter / Prospeo | Email discovery and verification | Before any bulk email send |
| ZeroBounce / NeverBounce | Deliverability verification | Before any bulk email send |
| Neon Postgres | Real database | Past a few thousand rows |
| A real CRM | Team collaboration | When more than one person is working the pipeline |
| A real data room tool | Permissioned document access | At first serious diligence |
If you move to a database or CRM, keep the CSVs as your source of truth and export from them. Two-way sync between files and a CRM goes wrong quietly.
Run your warm paths first. Build a small, well-qualified list. Send to 20 or 30, read the replies properly, fix the copy, then scale.
The single highest-leverage habit in the whole system is replying fast to anyone who shows interest. That one is entirely on you.
Good luck with the raise.
This guide covers version 1.0. Source: github.com/corebits-io/fundraising-team