Invention engine · law-based simulator · non-invasive brain uploading

Invent the ten missing links to a working brain–computer interface.

BCI v3 is a retrieval-grounded invention engine paired with a law-based simulator. It proposes designs for the ten hardest unsolved problems between a living brain and a verified digital twin — then grades every one against real biophysics, wave physics, information theory, and FDA safety limits. A design passes only if the physics permits it.

Launch the cockpit → View on GitHub
10
missing-link blockers
75 mm
non-invasive read depth
10¹⁴
synapses targeted
34
deterministic tests
Why this exists

Progress is blocked by ten problems that must be solved together.

A non-invasive interface that can map and emulate a whole brain is obstructed less by any single component than by whether a mutually-consistent set of components can satisfy every constraint at once — molecular delivery, deep readout, information capacity, data assembly, real-time simulation, and human safety. BCI v3 makes that search concrete: it invents candidate solutions, grades them against the laws that govern them, and fuses the ten passing designs into one end-to-end system.

The map

The ten blockers.

01
Pan-neuronal delivery
Label ~100% of neurons across the blood–brain barrier at a safe dose.
02
Multiplexed reporters
Give every synapse a demultiplexable molecular barcode in one voxel.
03
In-vivo readout at depth
Read barcodes non-destructively at 75 mm in a living brain.
04
Signal boost & noise cut
Keep detection alive at depth by brightness and averaging, not a bigger scanner.
05
Whole-brain throughput
Read the whole brain within a 30-day budget.
06
Trans-synaptic pairing
Recover which pre-barcode pairs with which post-barcode at F1 ≥ 0.9.
07
Exabyte assembly
Assemble the connectome within bounded storage at <5% error.
08
Real-time twin
Simulate 10¹⁴ synapses at a real-time factor ≥ 1.
09
Behavioural verification
Prove the twin behaves like the original at fidelity ≥ 0.9.
10
Human safety envelope
Hold ultrasound intensity, mechanical index, and viral dose inside FDA limits.
How it works

Invent → simulate → synthesize.

Retrieve. Search the literature (PubMed · arXiv · USPTO · Wikipedia) and ground each invention in real prior art.
Invent. Ten creative lenses propose designs under a strict non-invasive constraint — molecules and waves, never electrodes on brain.
Simulate. A deterministic law engine grades each design against biophysics, physics, electronics, and FDA ceilings. A persuasive narrative can never move the number — only the physics can.
Synthesize. When all ten blockers have a passing design, fuse them into one end-to-end brain-uploading system with a schematic, parts list, and interface budget.
The BCI v3 cockpit — pick a topic, invent, and watch the law simulator grade the design live
The cockpit: pick a blocker, invent with your local LLM, and watch the law simulator grade it live.
Invented designs graded by the law-based simulator
Every design is graded by the law engine — a pass means physically admissible, never proven in a living brain.
Ten solved blockers fused into one end-to-end brain-uploading system
Synthesize fuses the ten passing designs into one Label → Read → Map → Emulate system inside a safety envelope.
Prototype library

Mix & match, or build a curated prototype in one command.

Hand-pick which invention feeds each of the ten stages, or use one of six ready-made, internally-coherent presets. Every build is saved as a reusable prototype.

ECHO
Fully-acoustic coherent flagship — one readout physics end-to-end.
LUMEN
Optical / highest-resolution — fluorescence paradigm, 5 µm voxels.
SWIFT
Lean / fast — fewest reads, least averaging, max scan parallelism.
GUARDIAN
Clinical-safety-first — lowest intensity & dose, most rigorous verification.
TITAN
Maximum performance — highest raw scores across the board.
VANGUARD
Future / most-scalable — 2050-forward, novel-modality picks.
bci synthesize --preset echo      # or lumen | swift | guardian | titan | vanguard
bci synthesize --list-presets     # show all six
From prototype to paper

A research-grade monograph for every prototype — with a wet-lab plan.

Any saved prototype exports as a ~25–30 page journal-style paper: structured abstract, governing equations, per-blocker sections with computed figures, an integrated interface budget, a stage-gated translational roadmap, and a per-module wet-lab experiment program. Every number is a value the law simulator actually accepted — the document a lab would work from.

Research monograph — title, structured abstract, and prior-art/gap table
Journal-style monograph: structured abstract, prior-art gap analysis, governing equations.
Per-module wet-lab and validation experiments E1–E10 with model system, readout, and success criteria
Per-module wet-lab program (E1–E10): experiment, model system, primary readout, success criterion.
Run it

One command. Your machine. Your local LLM.

Local-LLM-first (Qwen via Ollama), with NVIDIA NIM and a rule-based fallback that needs no LLM at all. MongoDB for storage, JSONL when it's down.

pip install -e backend
bci serve                 # cockpit + API on one port → http://localhost:8000/app/
bci record multiplexed_reporters "acoustic, deep, safe" --constraint noninvasive
bci synthesize --preset echo
Get the code → Open the cockpit