AI you can
answer for.
One interface for your email, calendar, files, and client work — connected to AI on your machine. It shares nothing unless you say so.
curl -fsSL https://missioncanvas.ai/install.sh | sh
irm https://missioncanvas.ai/install.ps1 | iex
docker compose up
The model is always
the last step.
Most AI tools send your query straight to a model. MC sorts the work first. Then it gathers evidence, checks the rules, and only then calls a model.
Check the input
PII and PHI detected and stripped before classification. If the gate is unsure, it closes.
Sort the question
312-node ontology. Keyword → BM25 → LLM fallback. Every route gets a score.
Enforce the boundary
Sensitive domains block external calls. Not a config flag. A structural gate.
Gather the evidence
219+ knowledge entries. Freshness scored. Stale entries filtered before REASON.
Call the model last
Full structured context assembled first. Local or cloud — based on governance, not cost.
Work you cannot paste
into ChatGPT.
Legal
A federal court ruled in February 2026 that cloud AI can void privilege. MC routes legal work on your machine. That boundary is code, not a promise.
Healthcare
HIPAA 2026 takes effect August 2, 2026. PHI decisions need audit trails. MC records the path for every query — automatically, without configuration.
Financial services
Fiduciary work starts with the client's facts. MC pulls the client's files and your firm's knowledge before the model reasons.
Personal work
Write your doctor. MC uses your medical files and your tone. Health data stays on your machine. You do not configure a workflow.
Three things that meet at inference.
MC combines your professional context, your personal patterns, and your actual files at the moment of inference. The boundary decides what can leave your machine.
-
Canvas The rules for your field. Legal, healthcare, personal work. 25 bundled domains.
-
Lens How you work. Your tone, priorities, and sensitive zones. Grows from real use.
-
File map Your folders, not your file contents. MC learns where things live — what should stay put.
Every answer shows its work.
Every answer carries its route. You see the node, model, sources, and whether anything left your machine. No black box. No guessing.
Read the code that governs your data.
Mission Canvas is AGPL-3.0 open source. You can read every rule, route, and gate. You do not have to trust our word.
312 ontology nodes · 219+ knowledge entries · 2,506 tests passing
# No skipping. No shortcuts.
# The pipeline IS the OS.
def run(query):
query = scan(query) # gate: PII
node = classify(query) # ontology
path = traverse(node) # graph walk
kl = retrieve(node) # evidence
ctx = context(kl) # assemble
if node.blocks_external:
answer = reason_local(ctx) # ALWAYS
else:
answer = reason(ctx) # model last
store(answer) # path record: always
return answerFrom install to first answer
in five minutes.
Download MC
One download. No account. No email. A local model downloads. Your first canvas loads.
Ask once
The chips show the route, model, and whether anything left your PC.
Map your files
MC reads folder names, not file contents. Now it knows where your work lives.
Your lens grows
After 50 queries, MC knows your domains, patterns, and sensitive zones.
Use the portal
"Draft a response to Acme using the contract terms in my legal folder." One sentence. MC has the context.
Everything MC knows lives in files you can read, correct, or delete.
Free. Open source. Runs on your machine. No account.