I built Alice with local self-learning memory.
Alice is my outbound research, lead gen and email drafting agent.
She can crawl and enrich leads, store memory and RAG data locally in SQLite, generate drafts with a local LLM, create reports, and expose MCP tools for other AI clients.
But the real feature is this:
Alice does not write one email and forget.
When I accept, reject, send, or get a reply from a draft, she can store that outcome as memory. The next draft can use those lessons.
Every campaign makes the next campaign smarter.
Her sales voice is direct, evidence-led, and consultative.
No generic “hope you’re well” outreach.
Alice studies the business, finds a real angle, and frames the message around value creation.
She combines:
- lead research
- audit evidence
- brand voice rules
- accepted/rejected feedback
- outreach outcomes
- sales playbooks
- knowledge RAG
Every draft has intent:
warm first contact, value creation, objection follow-up, or closing.
The best part is how she angles things.
Alice can turn a small business weakness, unclear offer, or missed opportunity into a useful sales conversation without sounding pushy.
No cloud memory.
No black-box training.
Just local feedback loops, RAG, and measurable outcomes.
Less like a prompt.
More like a local operating system for outbound.