Behavioural AI coaching · now raising
Sanctum is a behavioural AI coaching platform. Every message runs through a six-agent cognitive pipeline that profiles personality, tracks behaviour over time, and calibrates every response to who the user actually is — delivered as one voice: Magistus.
The problem
Generic chat assistants give the same advice to everyone. They don’t remember last week, they can’t tell when you’re avoiding something, and they take what you say at face value — even when your behaviour says otherwise.
Real coaching works because a good coach builds a model of you over time: your patterns, your blind spots, the gap between your intentions and your follow-through. That’s the part nobody has productised.
Sanctum is built around exactly that gap.
How it works
The user only ever talks to Magistus. Behind every reply, six specialist agents analyse the message from different psychological angles. A synthesis agent resolves their disagreements and writes the single response — calibrated to the user’s personality and stage of change. No agent ever sees the full picture; that’s by design.
Runs first. Cross-references every message against the user's full behavioural history — recurring themes, avoidance patterns, growth edges, stage of change per topic.
Generates alternative paths and low-cost, reversible experiments — calibrated to how novelty-seeking the user actually is.
Pre-mortem analysis. Assumes the plan already failed and works backward. Owns commitment-load risk and cascade-failure detection.
Decomposes goals into ordered steps and finds the single first action that unblocks everything else.
Reads social capacity and energy state, then sets the voice the final response is written in.
Detects people-pleasing and the honesty gap — the distance between what the user says and what they do.
Inside the product
The user-facing response hides the machinery — but every agent decision is recorded and exposed. Pattern matches, risk assessments, personality scores, token costs, even the system prompts. This isn’t a mockup.
Why it’s different
Most coaching tools take users at their word. Sanctum's behavioural Shadow explicitly tracks stated goals against revealed priorities and surfaces the drift — gently, with evidence, not accusation. This is the core insight the whole system is built around.
The same risk is always identified. The same execution gap is always flagged. But how it's delivered changes per person — continuous Big Five (OCEAN) profiling tunes tone, structure, and framing to who the user actually is.
If a user is overcommitted, nothing new gets added until capacity is addressed. The system prevents the enthusiasm → overcommit → burnout → silence cycle before it starts.
Runs on self-hosted vLLM with open-weight Qwen3 models — not a thin wrapper on someone else's API. The intelligence, the data, and the model all stay under one roof.
The raise
Sanctum is already engineered — a production-grade backend, a full multi-agent cognitive pipeline, continuous personality profiling, and a working desktop and web client. What it needs now is capital to scale inference, complete productisation, and bring it to the people it’s built for.
This is a raise, not a sale. The founder retains ownership and direction. We’re looking for backers who want to fund the build and grow with it — not absorb it.
Move from development GPUs to a production self-hosted vLLM footprint that holds latency under real load.
Onboarding, billing, and the polish that turns a working system into a product people pay for.
Get Sanctum in front of the people who need a coach that actually models them.
Request the investor brief — architecture, status, roadmap, and the terms of the raise. Direct line to the founder.