AgentSource Plugin for AI Agents

B2B prospecting and company intelligence using the AgentSource API. Describe who you're looking for in plain English — your AI agent guides you through the workflow, shows you a preview, and exports to CSV.

Works with Claude Code, Claude Cowork, OpenClaw, and any other AI agent environment that supports skills and plugins.

How It Works

1. You describe a target in natural language
2. Claude (guided by the vibe-prospecting skill) parses your intent into API filters
3. It shows you market statistics and a data sample before touching any credits
4. You confirm — then it fetches your results and exports a downloadable CSV

All API responses are written to temp files so that large payloads never appear in the conversation context window, enabling scale queries of thousands of records.

Quick Start

1. Install

./setup.sh

2. Set your API key

Option A — paste it when prompted (Claude will ask on first use and save it automatically)

Option B — environment variable (recommended for persistent shell setups):

export EXPLORIUM_API_KEY=your_api_key_here
# Add to ~/.zshrc or ~/.bashrc to persist across sessions

Option C — CLI config command:

python3 ~/.agentsource/bin/agentsource.py config --api-key your_api_key_here

Don't have a key yet? Visit developers.explorium.ai for instructions.

3. Open Claude and describe your target

Find CTOs at Series B SaaS companies in California

Show me fintech companies using Stripe with 50–200 employees

Get emails for VP Sales at companies that recently raised Series A

Find companies hiring engineers that use Kubernetes

---

Requirements

• Python 3.8+ (standard library only — no pip required)
• An Explorium API key
• Any AI agent environment that supports skills/plugins (Claude Code, Cowork, OpenClaw, etc.)

---

File Structure

agentsource-plugin/
├── .claude-plugin/
│   └── plugin.json                  # Plugin metadata
├── skills/
│   └── vibe-prospecting/
│       ├── SKILL.md                 # Workflow logic — how Claude uses the API
│       └── references/
│           ├── filters.md           # All available filters with valid values
│           ├── enrichments.md       # Enrichment types and credit costs
│           └── events.md            # Business/prospect trigger event types
├── bin/
│   └── agentsource.py               # The CLI (pure Python stdlib)
├── setup.sh                         # Installation script
└── README.md

---

CLI Reference

All commands write results to /tmp/agentsource_<timestamp>_<command>.json and print only the file path to stdout. Use cat <path> or a python3 -c one-liner to inspect results.

API Key Config

python3 ~/.agentsource/bin/agentsource.py config --api-key <key>

Autocomplete (required for certain filter fields)

# Always pass --semantic for best results
python3 ~/.agentsource/bin/agentsource.py autocomplete \
  --entity-type businesses \
  --field linkedin_category --query "software" --semantic

Market Sizing (free — no credits)

python3 ~/.agentsource/bin/agentsource.py statistics \
  --entity-type businesses \
  --filters '{"company_size":{"values":["51-200"]}}'

Fetch (auto-paginates in batches of 500)

# Sample — 10 results
python3 ~/.agentsource/bin/agentsource.py fetch \
  --entity-type businesses \
  --filters '{"country_code":{"values":["us"]},"company_size":{"values":["51-200"]}}' \
  --limit 10

# Large query — 3,000 records (6 pages, automatic)
python3 ~/.agentsource/bin/agentsource.py fetch \
  --entity-type businesses \
  --filters '{"country_code":{"values":["us"]}}' \
  --limit 3000

Enrich

# Enrich companies with firmographics and tech stack
python3 ~/.agentsource/bin/agentsource.py enrich \
  --input-file /tmp/agentsource_xxx_fetch.json \
  --enrichments "firmographics,technographics"

# Enrich prospects with emails and phones
python3 ~/.agentsource/bin/agentsource.py enrich \
  --input-file /tmp/agentsource_xxx_fetch.json \
  --enrichments "contacts_information,profiles"

Events

python3 ~/.agentsource/bin/agentsource.py events \
  --input-file /tmp/agentsource_xxx_fetch.json \
  --event-types "new_funding_round,hiring_in_engineering_department" \
  --since "2025-01-01"

Export to CSV

python3 ~/.agentsource/bin/agentsource.py to-csv \
  --input-file /tmp/agentsource_xxx_fetch.json \
  --output ~/Downloads/results.csv

Entity Matching

# Match specific companies by name/domain
python3 ~/.agentsource/bin/agentsource.py match-business \
  --businesses '[{"name":"Salesforce","domain":"salesforce.com"}]'

# Match from an existing CSV
python3 ~/.agentsource/bin/agentsource.py match-business \
  --input-file /tmp/agentsource_xxx_from_csv.json \
  --column-map '{"Company":"name","Website":"domain"}'

# Match specific people
python3 ~/.agentsource/bin/agentsource.py match-prospect \
  --prospects '[{"full_name":"Jane Doe","company_name":"Acme Corp","email":"jane@acme.com"}]'

# Match prospects from a CSV
python3 ~/.agentsource/bin/agentsource.py match-prospect \
  --input-file /tmp/agentsource_xxx_from_csv.json \
  --column-map '{"Name":"full_name","Company":"company_name","LinkedIn":"linkedin","Email":"email"}'

CSV Import

# Convert an existing CSV for use with match-business or match-prospect
python3 ~/.agentsource/bin/agentsource.py from-csv \
  --input ~/Downloads/my_accounts.csv

---

Data Privacy

• API key stored at ~/.agentsource/config.json with permissions 600 (owner read-only)
• Also saved next to the CLI and in the current workspace directory for session persistence
• Temp result files at /tmp/agentsource_*.json — cleaned up by the OS automatically
• No data is ever written to the conversation context window

---

Troubleshooting

"EXPLORIUM_API_KEY is not set" → Run python3 ~/.agentsource/bin/agentsource.py config --api-key <key>, or export EXPLORIUM_API_KEY=your_key

"CLI not found" → Run ./setup.sh from the plugin directory

401 / Auth errors → Verify your API key at developers.explorium.ai

400 / Bad request errors → A filter value is invalid. Run autocomplete --semantic for the relevant field, or check skills/vibe-prospecting/references/filters.md

Large queries timing out → Reduce --limit or split the query by adding a narrower filter (country, company size, etc.)