Overview

AI services with their own curated knowledge, routed to the cheapest model that can answer.

Generic AI tools answer everyone's question from everyone's data. Kevin runs focused services for notes, meetings, research and device fleets. Each has its own knowledge base, and every request goes to the lowest-cost model that can handle it. You can see what each request cost.

Where to start Talk to us

What Kevin is

One platform, many focused services.

Kevin is a Python daemon with a set of services on top of it. Each service is narrow: one reads your documents and answers with citations, one transcribes meetings, one runs firmware on small radio-equipped boards. They share one platform, so a new service is mostly configuration rather than new code.

Behind the services sits an inference layer. It sends each query to a local model (through Ollama) or a hosted provider, picks by cost and capability, and records what the request cost. Data lives in SQLite by default, so there is no database server to run.

The family

Each member owns one job. Together they're the platform. See them all.

Operations and authoring
Marco
The browser-based surface for running the platform and writing in it: notes, tickets, specs and live system views.
Meeting intelligence
Saul
Transcribes and records meetings, and helps you recall and act on what was said.
RAG router
Annie
Sends each question to the right knowledge collection and the lowest-cost model that can answer it, with citations.
Web interface and monitoring
Mona
The React web app the daemon serves: dashboards, health checks and alerts in one place.
Edge firmware
Max
Firmware for ESP32-S3 boards with LoRa radios. WiFi, BLE and MQTT, updated over the air through a device registry.
Reactive keystore
Hank
A key-value store that tells subscribers when a value changes. SQLite by default.
Authentication and entitlements
Grant
Decides who can do what, and names the rule behind every denial.
Rules and compliance
Trixie
Turns policy and domain rules into checks on your content.

How a request flows

Three paths. The same bus.

Knowledge query

your question →
Annie picks a collection →
vector retrieval →
cost-routed model →
answer + citations

Telemetry stream

device →
MQTT →
Mx bus (latest values + history) →
Mona dashboard →
alert when a rule matches

Work dispatch

directive →
holon mailbox →
worker with the right skills →
result →
review

How it's organized

Four layers. They publish and subscribe on the bus instead of calling each other.

1. Infrastructure and roles
Vera (the edge server role), Grant (access), the message bus and the device registry.
2. Data pipeline and state
Flo, DynaRAG and the inference carousel move data through and keep track of current state.
3. Reasoning and inference
Rick, Annie, Max and Hank: retrieval, reasoning, edge firmware and the keystore.
4. Interface and collaboration
Marco, Noted, Norm and the knowledge graph: where people actually work.

Where to start

Three doors, depending on what you need.

Try it on your documents

Vera indexes a folder of files on your own machine and answers questions about them, with citations.

Meet Vera →

Run the platform yourself

Self-host the daemon and the services you want. See what is packaged today and what still runs from source.

See downloads →

Talk to us

Tell us what you want to build. We'll say honestly whether Kevin fits.

Get in touch →