Skip to content

Repository files navigation

Prospecting Agent

Finds your next customers, researches meeting notes, writes the email, sends it, and handles the reply.

Runs on your laptop, and compatible with DGX spark (see dgx-model-serve) for local inference.

Configure the agent to gain access to:

  • Apollo — finds contacts and companies, fills in what's missing
  • Tavily — reads up on the account and catches buying signals
  • Your local LLM — writes every email (OpenAI-compatible; gpt-oss-120B here)
  • Gmail or SendGrid — sends it, then watches for the reply
  • HubSpot or a CSV folder — bring a CRM, or don't

All you have to do is bring your API keys, fill in .env, and run it!


Table of contents

  1. A day of it
  2. Quickstart
  3. Integrations
  4. Docs
  5. License

A day of it


The worker keeps its own schedule. You do not start anything by hand.

Time (UTC) What happens
06:00 New contacts in data/ are imported
06:30 / 06:40 Contacts and companies enriched
06:45 New contacts discovered at target accounts
07:00 Companies researched
07:15 Contacts enrolled in an outreach sequence
08:00 Buying signals detected
hourly Email sent — only inside your cadence windows
hourly Replies read and classified, follow-up tasks run
16:30 Daily digest, after the send window closes
Mon 09:00 Weekly report

Nothing is sent until you say so — DRY_RUN=true is the default. Apollo spend sits under a daily credit cap you set, with a per-run budget on top of it.

Quickstart


./scripts/setup.sh                              # deps, services, build, tests
# fill in your API keys in .env
prospecting-agent gmail-auth --daily-limit 25   # authorize a sending account
prospecting-agent health                        # db, LLM, memory, gmail, capacity
prospecting-agent worker                        # run it

prospecting-agent serve adds a local control API for running a job now, checking a contact, or posting an inbound reply. See Operations.

Integrations


Integration Role Required
Apollo Contact and company discovery + enrichment Yes
Tavily Web research and buying-signal detection Yes
Local LLM (OpenAI-compatible) All generation and analysis (e.g. gpt-oss-120B) Yes
Mem0 (self-hosted) Semantic memory over research and interactions Yes (bundled in compose)
Postgres System of record + job queues Yes (bundled in compose)
Gmail Email sending + reply monitoring One email provider
SendGrid Email sending (alternative to Gmail) One email provider
HubSpot CRM sync + engagement logging Optional (CSV works without)
CSV import No-CRM contact source Optional
Slack Rep notifications, digests, error alerts Planned (M6)
HeyReach LinkedIn outreach Optional (off by default)

Docs


License


MIT: LICENSE.

About

Open-source AI prospecting agent: use me for lead discovery, research, and multi-channel outreach automation

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages