About
Cognitive science as the blueprint, and a standing rule that a claim ships only after it survives production. Three parts: the company, Labs, and Applied.
SMTRY builds ownership-first AI software: Anamnesis, live in alpha, and Atria, in development. SMTRY Applied runs engagements on that same stack, plus select outside projects. SMTRY Labs is where the research happens, including the Craft alpha.
We started as Project Symmetry. We shortened the name. The premise stayed: AI software a person or a firm owns outright.
J. I. Ashley Consulting is our sibling firm for business-focused advisory: strategy, including AI strategy, program delivery, brand and marketing, web design and operations, and AI search visibility. Neither firm builds generic custom software, and when a client's work belongs with the other firm, we say so and hand it across by name.
Methodology
What we sell, and what we don't
The architecture starts from a question about the firm rather than the model: what does it need to know, how confidently, and what happens when it is wrong? That is an epistemological problem before it is a technical one, and the engineering follows from the answer.
Define the knowledge requirement first, build the system to meet it, and challenge every output before trusting it. Parameters that perform earn the right to stay. The rest is discipline: a conviction formed without evidence is an opinion, and we do not ship opinions.
SMTRY builds and sells software a person or a firm owns outright. Strategy, AI visibility measurement, program delivery, and advice are the work of J. I. Ashley Consulting, the sibling firm that built SMTRY. Each refers the other's work across by name, and neither builds generic custom software.
SMTRY Labs · research
Labs is where we test claims before they go on a product page.
The question was whether memory can run entirely inside a network, on a small model, and still be trusted. In August 2026 we ran a holdout: an 8-billion-parameter open-weight model, quantized to 2 bits, against a cloud arm. Evidence coverage was 0.94 on both arms. The local arm reached 17 of the cloud arm's 22 correct answers. It took about 70 hours where the cloud took 6 minutes. So the local model can do the work, and it's slow. The part that's fast: local screening matches frontier verdicts at 0.95 confidence in 2 to 3 seconds. That's the piece in use today: screening before trust.
August 2026 holdout · local 8B 2-bit vs cloud pipeline
Craft Alpha
We built the memory because we needed it, and shipped Anamnesis first. The measurement came after, once the question was whether it could run inside a locked network. That's the right order for a product and the wrong order for a paper. We'd rather publish the numbers late than the claims early.
Craft is an alpha we run for our own writing desk rather than sell: a collaborator that learns a writer's instincts and pushes back on the places where habit is standing in for a choice. It edits, it argues, and it keeps a record of what it changed and why. It is not available outside the firm.
The measured results and their methods are published on the product's benchmarks page and kept current; every figure on this site links there.
SMTRY Applied · engagements
For builders whose data can't leave the building, or must be auditable and deletable inside it.
Warm recall on live accounts. Every engagement below is this product, in your environment.
Evidence coverage, local open-weight against the cloud pipeline, August 2026 holdout.
Claude Code, Claude Desktop, claude.ai, Cowork, ChatGPT, Codex CLI, Gemini CLI, one connector.
Engagements
Anamnesis connected to the tools your team uses, running against your accounts, with capture and recall verified on your work rather than ours. Priced as a fixed fee scaled to the seats involved, so it reads as a product purchase with setup.
The memory stack installed inside your network, keys held by you, retrieval and screening run on hardware you control. You get the same visible, searchable, deletable memory, on your side of the wall. Priced as a fixed fee scaled to the hardware it runs on, plus a support retainer.
Custom builds that extend the stack, a new connector, an Atria function, an evaluation harness, are scoped by conversation and never listed.
What Applied refuses
Strategy, measurement, and advice
Staff augmentation. Hourly work. Generic applications. Anything that is not on SMTRY's own stack, unless it is a project we have selected for a reason we can name. Strategy, brand and marketing, web design and operations, and AI search visibility are J. I. Ashley Consulting's work, and we send them there by name.
If your question is whether to adopt memory at all, what to measure it against, or how it fits a program you're already delivering, that's advisory, and it goes to J. I. Ashley Consulting. They do strategy, including AI strategy, and program delivery. When they find a memory problem that needs building, they send it here.
SMTRY