Your AI has Amnesia.
Dr. MnemosyneC has the Cure. Substrate.
The Substrate Cure to AI Amnesia.
Every time you start a new session, your AI forgets everything. Your projects, your preferences, your past conversations, gone. Dr. MnemosyneC gives your AI a permanent, private memory that actually stays.
Mac / Linux - coming soon
Works alongside ChatGPT, Claude, Gemini, or any AI you already use.
All your data stays on your own computer. No cloud. No account required.
FREE (SSPL license) to use FOREVER. No Ads. No Strings.
Optionally join the Cooperative Universal Substrate™ for $5 a year. Not required.
What your first session looks like
- Tell it something. “I’m working on a report about water treatment costs in rural areas.”
- Come back three days later. Start a fresh session. Ask about it. It remembers.
No copy-paste. No re-explaining. Pick up exactly where you left off, every time.
Free. Or pair MnemosyneC with a paid AI like Claude or GPT for faster, bigger answers. Either way, the substrate carries the memory, so your paid AI bill stays smaller too.
Just Add Salt. How to Get the Right Answer.
The cooperative-class architecture for local-first AI that works at 97.1% on the canonical 70-question MMLU-Pro benchmark · on consumer hardware · with no paid API keys · empirically reproducible.
Does It Actually Work?
PROVE IT YOURSELF75 questions · 4 AI vendors · real test run 2026-05-30. No tricks.
The Banyan Metric measures recall of substrate-stored information. “Without” = model alone, cold start. “With” = same model after MnemosyneC substrate context injected.
The free local model (Ollama, $0 inference cost) lifted from 6% to 78% entirely because of the shared substrate. Full proofs →
📌 Pinned Proofs
Plain receipts. Caveats stated aloud. No hype. Click any row to expand.
Storm Test Three failures at once. Nothing lost. ✓ Zero work lost
A routine update wiped connectors. A hook-script deletion blocked every prompt. The AI service returned overload errors. Three simultaneous infrastructure failures. The substrate held.
✓ Zero work lost
A real tornado warning. Then three systems broke at once.
System broken underneath. Work continued anyway.
Tally confirmed. Everything recovered. Zero work lost.
Receipt: BP071 close stamp →
Mesh Proof A node answered for data it did not hold. ✓ 20/20 · 16.6 ms p50
Machine A had never seen the data. Machine B held it. Connected on local network, Machine A answered 20 of 20 correctly, hash-verified. Median response: 16.6 ms.
✓ 20/20 · 16.6 ms p50
Receipt: Asteroid-ProofVault/MESH_6_RECEIPT_BP063.md →
Benchmark R10 A free, local Ollama model jumped from 6% to 78% accuracy. ✓ +72–83 pts
A free, local Ollama model jumped from 6% to 78% accuracy. No retraining. Just substrate.
Cohen’s κ = 0.936. Three independent raters. Hash-verified. Caithedral™ Core applied substrate context — no model fine-tuning, no new training data. The substrate did it. (+72–83 percentage point gain, BP065)
Receipt: Benchmark R10 · BP065 →
Trial 02b 4-peer cooperative gemma4:12b · 81.9% active-peer aggregate via relay mesh. ✓ 172/210
Pass A anchor: 70/70 claude-sonnet-4-6 on M0. Pearl: 0fa461c8.
Pass B 4-peer via relay: 70 questions dispatched to all 4 peers via relay.lianabanyan.com (LAN-AS-WAN constraint honored). Model: gemma4:12b on each peer.
| Peer | Score | Status |
|---|---|---|
| M0 (cb4ef450) | 57/70 (81.4%) | Active |
| M3 (d0b47bd0) | 2/70 | Offline |
| M2 (88cbf6bd) | 61/70 (87.1%) | Active |
| SON (49f3e597) | 54/70 (77.1%) | Active |
Active-peer aggregate (M0+M2+SON): 172/210 = 81.9% Full 4-peer aggregate: 174/280 = 62.1%
Variance: 70/70 full agreement — every peer that answered agreed on every question.
Pass A pearl: 0fa461c8 · Pass B pearl: a7a290f2424398e5 · BP089 · 2026-06-21
HOW IT WORKS — The Substrate Persistent Memory Is the Asset · How We Make Sure Things Are True & Quickly & Cheaply Accessed.
That free local model answers at $0 a call. With our Patented Substrace Theorem, you will get the correct answer, in time, and if you or anyone else in the Cooperative Substrate ever asked it before - you'll get the answer FAST and FREE. The Mesh Frontier™ runs on hardware every user already owns. No new data center needed, since this makes existing data centers 95% cheaper, more efficient, AND more accurate! A third option to the Data Center Dilemma™.
MnemosyneC has three layers. Each does one thing well. Click any row to expand.
Reader — Answers your question — Gemma 4 12B (local, free, private) OR ADD your Paid AI of Choice.
We basically add a virtual Library of Alexandria — where the data you approve (answers) get copied for everyone to read, and you keep yours. And get access to theirs.
Verifier — Adjudicates whether the answer is correct — Shadow E-Giant™ concordance (3+ parallel votes)
This combines with our Plow Loop™ process to loop as many times as it takes — at $0 per call — to get the right answer and STORE it for FAST ACCESS EVER AGAIN. Or, you can use your Paid AI of Choice to do it faster. Either way, the Plow Loop makes it 100% accurate. And faster. And Cheaper. And Stored. And Shared. And Shared WITH.
Accumulator — Stores verified knowledge for next time — Eblet store (append-only JSONL, SHA256-stamped)
A Vault of Anonymous Answers™, so you can save on what others have paid to find — in return for providing the same.
Use Ask in MnemosyneC. Use Claude Desktop. Use Cursor. Use ChatGPT in your browser (browser extension coming v0.2). Whichever surface you use, the substrate underneath compounds. Your knowledge grows. The model is interchangeable. The substrate is what makes any AI smarter — for you, for your work, for the keep™.
The substrate doesn’t just store knowledge. It lives.
Pheromones decay.
Every claim entering the substrate starts as a pheromone — a signal that something might be true. Pheromones have a half-life. If nothing refreshes them, they fade. Old guesses die naturally. The substrate doesn’t accumulate noise — it lets noise wither.
Socceri triads confirm.
When three independent verifications converge — three Shadow E-Giants from different perspective lenses (correctness · consistency · coverage) returning the same verdict — a pheromone becomes a Socceri Triad™ confirmation. Verified. But still subject to decay. Verification without continued use still fades.
Living connection buoys.
Confirmed claims survive only if the community keeps them alive — through use, through query, through federation. Every time another user’s substrate concordance-aligns with yours, the claim is buoyed. Without that living connection, even verified truth withers on the vine.
Stone tablets are forever.
When a claim has been buoyed by enough living connection over enough time, it gets promoted to a Stone Tablet™ — immutable, append-only, permanent. Stone tablets are the substrate’s load-bearing canon. They don’t decay. They are the truth the network has earned together.
This is the innovation itself.
Truth isn’t asserted by authority. It’s earned through use, hardened through verification, and preserved through community.
- Failed claims die. No noise accumulates.
- Verified claims persist if used. No abandoned facts survive.
- Used-and-verified claims become permanent. Stone tablets compound the network’s hardest-won knowledge.
The substrate gets cleaner over time, not noisier. It’s how biological memory works. It’s how scientific consensus works. It’s how cooperative-class™ trust always worked, before extraction-class platforms tried to centralize it.
We’re just naming the mechanism, building it into code, and giving you the keys.
Your Frame Is One Node of the Frontier Mesh.
MnemosyneC is not a chatbot. It is not a retrieval system. It is not RAG. RAG fetches from a static store. The substrate is a living cooperative mesh -- every verified answer strengthens the mesh for every node that shares it.
Your machine is a node. Every correct answer you add is available to every peer you choose to federate with. Every correct answer they add is available to you. The mesh is the knowledge. The node is you.
This is why a free local model (Ollama, $0/call) can jump from 6% to 78% accuracy. Not because the model improved. Because the mesh did.
RAG gives you a library. The Frontier Mesh gives you a cooperative.
Substrate Works Without MnemosyneC Running.
Bring your own AI. The substrate is a flat file on your machine. Any AI can read it. MnemosyneC is the interface -- not the dependency.
Three tiers, all supported:
- Free local (Gemma / Llama / Mistral) + substrate -- $0/call, private, runs offline, substrate compounds with every session
- Flagship (Claude / GPT / Gemini) + substrate -- faster answers, same substrate compounding, lower cost per session over time
- Standalone substrate API (no AI model) -- query the substrate directly, build your own integrations, SSPL licensed
Flagship WITH Substrate = BROKE THE SOUND BARRIER
You do not need an Anthropic API key. You do not need a Google API key. You do not need anything but the substrate and a model that can read it. We just make it easier.
Here's Why This Is a Good Deal for You.
If you are an AI company, an enterprise AI buyer, or a developer integrating language models, the substrate is not a threat. It is a competitive advantage.
- Cost: Substrate-cached answers cost a fraction of cold inference. Less compute per useful answer.
- Throughput: Verified answers resolve in milliseconds from local store. Your model handles the hard cases, not the already-answered ones.
- Accuracy: Shadow E-Giant concordance filters out hallucinations before they enter the substrate. Quality compounds -- it does not degrade.
- Vendor resilience: The substrate is model-agnostic. When you swap models, the substrate stays. No knowledge loss at model transition.
- Brand-defensive: Users who trust a cooperative substrate trust the AI that uses it. The cooperative model builds durable trust that extraction-class platforms cannot replicate.
Cooperative Defensive Patent Pledge #2260 is the legal mechanism: members and the public can use the substrate architecture. This is not mercy. This is structural. We win when the substrate grows. You win when you build on it.
Direct download from Liana Banyan Corporation. SHA-512 verified.
One quick thing before you run the installer
When Windows sees new software for the first time, it shows a caution screen. This happens with all new programs, including ours. It does not mean there is a problem. When you see it, click More info and then Run anyway to continue. The installer is safe.
- Click the download button above and save the file. (Large file: the bundled AI model is included.)
- Right-click the downloaded file → Run as Administrator.
- When the Windows caution screen appears, click More info.
- Click Run anyway. Follow the installer prompts.
- Done. Your substrate starts the moment you do.
Version 0.8.29 · Free forever · No account required · All data stays on your computer
About Liana Banyan
Liana Banyan is the Capitalist Cooperative. We do it fairly, for the benefit of all. For it rains on the just and the unjust.
Liana Banyan Corporation — the cooperative-class substrate behind MnemosyneC and 16 cooperative initiatives.
83.3% of every dollar stays with the Workers, Builders, and Creators that make it. $5 a year. One vote. Everything we build is published under a Cooperative Defensive Patent Pledge so members — and any community on Earth — can fork it, run it, and keep it.
Read the papers that shape the work: How to Save the World in Six Easy Steps · A Considered Approach to Universal Sustained Economic Prosperity · No Atomo. Superman! · The Invisible Tax
Read the Cooperative Compact → · Browse initiatives → · Join for $5/year → · The Court →
Caithedral™ · Substrate · Plow · Stone Tablets · Shadow E-Giants · Cooperative-class · For the keep.