01
Character as product
Voice and behavior were specified, tested, and guarded as core functionality rather than decorative prompting.
CASE / 03 · Local AI · designed personality
Omni Mind
AI Companion
A lore-aware AI companion designed to stay in character, remember meaningful context, operate with moderation guardrails, and run on the client's own hardware.
Local LLMFine-tuningMemoryGuardrailsDiscord
Local
Runs on client-controlled hardware
In character
Personality designed as a system
Discord-native
Delivered where the community already is
Screens withheld · client request
The challenge
A convincing companion needs more than a system prompt. It has to know the world's lore, preserve a recognizable voice, remember useful context, handle unsafe interactions, and respond inside the community without repeatedly stepping outside the fiction. The client also wanted the system to operate on hardware they controlled.
What we built
Windrose developed a local model pipeline around the lore corpus and wrapped it in a deliberately designed personality. Persistent memory, moderation rules, retrieval, and response controls were integrated with Discord so the companion could live where players already gathered. The deployment path kept the model and player data on the client's infrastructure while giving the team direct operational control.
LoreA structured corpus and retrieval path that gives the model the right world knowledge at the right moment.
CharacterPersonality rules, examples, fine-tuning, and response constraints designed to resist breaking character.
MemoryPersistent context selected around meaningful interactions without turning every conversation into an unlimited transcript.
CommunityDiscord-native interaction, moderation guardrails, operational controls, and local deployment.
A companion that feels part of the world.
The delivered system combined local inference, lore grounding, personality, memory, safety controls, and community integration into one reliable experience. The technology stays behind the character instead of constantly reminding players they are talking to a generic bot.
/02 — Engineering decisions
The choices beneath the interface.
The strongest AI product decisions are often invisible to the user. These were the boundaries that kept the system coherent, maintainable, and ready for real use.
02
Local by design
Deployment and model choices were shaped around client-controlled hardware and private conversation data.
03
Selective memory
The system retains useful context while keeping memory bounded, relevant, and operationally manageable.
04
Native delivery
Discord integration made the companion part of the community instead of requiring a separate destination.
The Omni Mind AI bot demanded both technical depth and a strong sense of character. Windrose handled the complexity thoughtfully, communicated clearly throughout the build, and delivered a system that felt reliable, distinctive, and ready for its community.
/03 — Continue exploring
Related work + capabilities.
Explore the capabilities behind this product and how we approach a focused MVP launch.
Your MVP.
Live in 30 days.
We get your MVP live in 30 days or you don't pay. Tell us your idea and who it is for. We will reply within one business day to help shape your launch plan.
Let's launch your MVP ↗Scope, launch criteria, and kickoff agreed before the build. How the guarantee works.