01 — Abstract
Bad Theory Labs is a research lab and product studio building the layer that makes AI useful inside real work: memory, retrieval, context, and presence. Today's models can talk; they cannot reliably remember, follow through, or stay grounded. We treat that as an infrastructure problem, not a prompting one.
This document describes the thesis behind the lab, the products shipping against it, and $BTL: a Solana token that earns holders discounted credits for BTL Runtime, gates early access to what we build, and lets the community coordinate around it. $BTL is a utility and access asset — not a promise of profit.
02 — The problem
Frontier models are extraordinary pattern completers and unreliable colleagues. They hallucinate facts, forget what you told them five minutes ago, and re-derive the same context every session. The industry's answer has mostly been “bigger model, longer prompt.” That scales cost, not trust.
- No durable memory. Context windows are not memory; they evaporate.
- Hallucination under retrieval. Most RAG stacks confidently invent when the answer isn't present.
- No presence. Models react to a turn; they don't persist, watch, or follow through on work.
03 — Thesis
Intelligence is compression. To perceive is to compress input into representations that keep what is useful. To reason is to compress those further into structure that supports prediction and transfer. We believe the durable advantages in AI will come from systems engineered around that principle — grounded memory and causally-structured retrieval — rather than from raw scale alone.
The field has been teaching systems to predict the world. We are building systems that hold onto it.
The research backing this lives in our Compression Program; the products below are its commercial edge.
04 — The stack
RetainDB
Grounded memory + retrieval that refuses to invent. 0% hallucination on a frozen dataset and 79% on LongMemEval versus 56.7% for the next best system.
Marrow
The presence layer — gives an agent a persistent working context so it can watch, remember, and follow through across sessions instead of starting cold every time.
BTL Runtime
A model gateway that routes, meters, and hardens traffic across providers — the plumbing that lets teams run agents in production without rebuilding it themselves.
BTL-2 Coder
A compact 7B coding model fine-tuned in-house — our testbed for compression-first training and the first model carrying lab methodology end to end.
05 — The $BTL token
$BTL is an SPL token on Solana. It exists to align incentives between the people building the lab and the community supporting it. Its design goals are deliberately narrow:
- Cheaper credits. Holding $BTL earns discounted credits for BTL Runtime access — the more you hold, the less you pay per call.
- Early access. Token-gated entry to product betas (RetainDB, Marrow) and priority hackathon tracks.
- Coordination. Signal weight in community proposals — what we prioritize, what we open-source.
- Membership. A durable badge of being early to the thesis, redeemable for community standing.
What $BTL is not: it is not equity, not a share of revenue, not a debt instrument, and not a promise that anyone will profit. Utility ships incrementally and is documented publicly as it lands.
3bBQrzzq9DRXXFfC9nUno9m1MBm9Y7dVnBBK44bVpump06 — Distribution
$BTL launched into a fair, open market — no insider pre-sale round. The intent of the distribution is broad community ownership and a treasury that funds research and hackathons.
Indicative allocation. Final on-chain figures are verifiable on Solana explorers via the contract address above; this section is descriptive, not a binding commitment.
07 — Roadmap
08 — Risks & disclaimers
$BTL is a high-risk digital asset. Crypto tokens can lose all value; only participate with funds you can afford to lose entirely. Nothing in this document is financial, investment, legal, or tax advice, and nothing here is an offer or solicitation to buy or sell any asset. $BTL confers no ownership, equity, dividend, or revenue right in Bad Theory Labs or its products. Roadmap items are aspirations, not guarantees, and may change or be dropped. Always verify the contract address yourself before transacting. Regulations vary by jurisdiction — it is your responsibility to comply with yours.