DeepFounder // AI LABORATORY Book a demo
EXP-04 · spring 2026 DF-LAB / NEW YORK · LISBON

An AI laboratory
building the operating system
for small SMB · US business.

DeepFounder researches and ships proprietary language models, autonomous agents, and business intelligence for the operators who keep the U.S. economy running. One lab. Three products. Built for owners and operators — not engineers.

Book a demo → Read about the lab ↗
// Status01
Online/ all systems
// Active runs02
1,247/ ops
// Coverage03
United States/ SMB market
// Lab04
DF-Labs/ EXP-04
01 — Products in production

Three systems.
One laboratory.

Each product is a research artifact in active deployment — calibrated for the texture of small business work, then hardened in real operations.

PRODUCT — 01
Self-hosted AI agent

qwe-qwe

DF-AGENT-1 · v0.18.5 · MIT · Python 3.11+

A business-oriented AI agent that drops into real workflows — customer ops, internal automation, knowledge retrieval, scheduled reporting. Self-hosted on your infrastructure. Bring any OpenAI-compatible LLM — Azure, Bedrock, OpenAI, Groq, OpenRouter, or a local model on your own hardware. Your data, your provider, your rules.

  • runtimeSelf-hosted · laptop, server, or cloud
  • llmAny OpenAI-compatible · 7 providers
  • stack46 tools · 7 skills · MCP · cron
  • accessWeb UI · CLI · Telegram
View on GitHub →
FIG. 01 / AGENT TOPOLOGY DF-AGENT-1 8 core tools · tool_search v0.18.5
PRODUCT — 02
AI business intelligence

DF Insights

DF-BI-1 · 15 cities · 85,560 businesses

AI-powered competitive intelligence for U.S. local businesses. Deep-learning models analyze digital presence, reviews and market positioning — then ship a per-business report with competitor intel and growth recommendations. 85,560 reports already generated across 15 cities and 9 categories.

  • marketU.S. local businesses · 15 cities
  • analysisDigital presence · reviews · positioning
  • outputPer-business report + competitor map
  • scale85,560 reports · 100% coverage
Explore reports →
FIG. 02 / COVERAGE DF-BI-1 15 cities · 85,560 reports 2026.05
PRODUCT — 03
Proprietary models

DF Foundation

DF-LM family · trained in-house

Our own family of language models, trained from the ground up on the language of small business — invoices, schedules, vendor contracts, support transcripts. Smaller, faster, and meaningfully better at the operator's job than general-purpose frontier models.

  • familyOperate · Insight · Reason · Voice
  • trainingCurated SMB corpus, in-house
  • deployAPI · on-prem · edge
  • evalBeats GPT-class on SMB tasks
Read research notes →
FIG. 03 / EVAL DF-LM family SMB-bench v3 n=4820
02 — Field notes

Latest from the lab.

Research notes, field reports and methodology pieces from the team — published openly on https://deepfounder.ai/blog.

Castor Jun 05, 2026

Castor for Pediatric Clinics: Offline AI That Documents Well-Child Visits and Immunization Records Without Cloud Risk

Offline AI for pediatric practices: well-child visit notes, immunization records, EPSDT screenings, and prior authorizations — all running locally with no cloud exposure.

Kir Leshkevich · 6 min read Read →
Castor Jun 04, 2026

Castor for Home Health Agencies: Offline AI That Documents Every Patient Visit Without Cloud Risk

Home health agencies collect HIPAA-protected patient data at the bedside every day. Castor runs fully offline — no cloud, no PHI exposure — handling EVV compliance, SOAP notes, medication records, and Telegram-based care coordinator alerts.

Kir Leshkevich · 7 min read Read →
Castor Jun 03, 2026

Castor for Addiction Treatment Centers: Offline AI That Keeps SUD Records Off the Cloud

Keep SUD records off the cloud with Castor offline AI. SOAP notes, MAT tracking, UA docs — 42 CFR Part 2-safe on your own hardware.

Kir Leshkevich · 11 min read Read →
// archive · since 2024 Browse the full archive →
03 — Work with the lab

Run a small business?
Let's build something.

We onboard a limited cohort of operators each quarter. Tell us about your business and which product fits — Qwe-Qwe, DF Insights, or both — and we'll schedule a 30-minute working session within five business days.

EMAIL [email protected]
OFFICES NEW YORK · LISBON
NEXT COHORT Q3 2026 · 12 OPERATORS
// Namerequired
// Work emailrequired
// Business+ industry
// Interested inselect one
// What you want to solve1–2 lines