An AI that actually knows your CRM

The problem

Every CRM vendor shipped a chatbot in 2024. Almost none of them can answer a question about your data — they summarize help docs, not your donors.

How Chonk handles it

• Ask your CRM (RAG) — natural-language questions grounded in your actual records, with the contacts attached to the answer.
• Semantic search — find people and notes by meaning, not just keyword match.
• Donor RFM scoring — rule-based, deterministic, and always on (no AI config required for this one).
• AI insights & email drafting — surface patterns and draft outreach in your voice.
• MCP tools — ask_crm, semantic_search, and donor_score are exposed over the Model Context Protocol, so agents like Claude can do work in Chonk.

How it’s wired

The AI layer talks to any OpenAI-compatible provider and is configured entirely by environment variables — swap models without touching code. Embeddings are an optional separate endpoint: turn it on and semantic search uses true pgvector; leave it off and Chonk gracefully falls back to Postgres full-text search. The rule-based donor score works no matter what.

The flex

Because the tools are MCP-native, Chonk isn’t just AI-assisted — it’s AI-operable. An agent can look up a donor, score them, and draft the ask, all through typed tools instead of screen-scraping.