LLM features that do real work for your users: assistants, retrieval over your own data and automated workflows. We build in guardrails and evals.
AI features
We treat the model as one component in a system: retrieval over your data, memory, streaming, voice, and a privacy layer that strips personal details before anything leaves your servers.
What you get
Oslyn, Brologue's AI coach: an agent with memory, tools, approvals and voice.
An agent with memory, retrieval over the user's journal, photos, and tools across the app and connected accounts. Anything that acts outside the app, like sending an email, waits for the user's approval.
Talk to Oslyn out loud: Whisper transcribes, ElevenLabs Flash speaks, and Groq's Orpheus takes over if ElevenLabs is unavailable. Meditation audio comes from Kokoro TTS deployments we run ourselves.
How Oslyn handles a message
Each turn, a fast model classifies the message, which decides the tools and memory Oslyn gets. Reading is free; anything that acts for the user outside the app waits for their approval.
Message
Text, a photo, or speech transcribed by Whisper
Classify
A fast model reads the intent and picks the tools and memory
Context
Profile, long-term memory and journal retrieval via pgvector
Act
Tools read sleep, tasks and calendar, and connected apps
Approve
Emails, calendar events and task changes wait for a yes
Stream
Tokens to the app, or on to ElevenLabs as speech
A 30-minute call about the problem, who uses the product, and what a good outcome looks like.
A written scope with milestones, a timeline and a fixed price for each milestone. Nothing is billed until you agree to it.
We ship working software every week. You can follow progress in the client portal and try each build as it lands.
We deploy the product, hand over the code and docs, and stay on call while real users start using it.
Thirty minutes with an engineer. Bring a deck, a Figma file or just the idea. You'll leave with a clear next step either way.
Free · 30 min · Times in your time zone