
Your private space to think freely.
Protect ideas, messages, questions and private plans. Local-first and under your control.
Private on-device beta. Generative AI is not connected. Use FINCH Fusion to inspect model routing. No prompts are transmitted.
Control what leaves your device.
Inspect data locally, classify risk, and redact sensitive material. No text is uploaded during scans.
Privacy inspection
Disclosure-safe preview
Enforced boundaries
Pattern matching is not comprehensive DLP. Do not put live credentials or regulated records into a beta.
Where can this data go?
Evaluate a disclosure boundary before routing information. The default is local-only. Remote routes are denied until verified CPX-controlled services exist. Use synthetic samples in this beta. Personal and Business cryptographic compartments are separate.
- Data boundary
- Pattern classes
- Retention posture
This local heuristic engine is not certified DLP and does not prove that text is safe. Receipts contain no raw prompt, but metadata can still be sensitive.
Test. Compare. Improve.
Same 24 synthetic checks, three independent devices, one reproducible baseline.
Run the test suite
Three-device correlation
Each tester sends only the exported synthetic result JSON. Import them together here; no server is involved.
One intelligence layer. Every useful capability.
OpenAI-first reasoning, optional external model challengers, verifiable search and private analysis. FINCH selects the right route based on privacy, evidence needs, cost and availability. No silent external transmission.
Route simulator · real local policy
Inspect what FINCH would do before any provider sees a prompt. This is a routing decision, not generated model output.
Model adapter registry
Only FINCH Local is currently active. No third-party API token is present in the browser or this beta. Provider privacy policies and account terms must be verified before enabling live routes.
Own the intelligence, not just the interface
The proprietary layer is the policy router, evidence graph, evaluation harness, encrypted user memory, allowed tool execution and user-authorized training corpus. External models remain replaceable until a FINCH-owned model meets independent benchmark thresholds. Do not train on third-party model outputs without license authorization.
Your FINCH. Your rules.
Direct control of local history, AI routing, and beta readiness.
Encrypted conversation vault
Your passphrase is not transmitted or recoverable. Unlocking makes history readable on this device only.
Model & access posture
On Android, use Chrome's menu → Add to Home screen → Install if available.