SMTRY

Products

Anamnesis first.
Then what we can measure.

SMTRY sells software a person or a firm owns outright. One product is live. The second is named and not yet described, by rule: a sentence about a product says what a buyer receives.

Anamnesis Live · free alpha

One memory, read by every AI tool you use.

Say a thing once, in any of seven tools, and it is there in the others: Claude Code, Claude Desktop, claude.ai, Cowork, ChatGPT, Codex CLI, Gemini CLI. Each account holds its own keys; there is no master key. Export is plain JSON, anytime. Warm recall takes 0.2 to 0.4 seconds.

Try the alphaHow it protects you →

The Anamnesis dashboard: tokens saved, tokens paid, retrievals, crystallizations and memories
Your account's memories as you see them: all of them, searchable, deletable. The benchmarks → · Docs →

Architecture

One memory pipeline, grounded in memory science.

One memory layer, seven clients on top of it, your keys underneath. Anamnesis is the product; the plugin sits in Claude Code, Claude Desktop, claude.ai, Cowork, ChatGPT, Codex CLI, and Gemini CLI, and all seven read the same store. What you tell one tool is available in the others. Below the store: per-account encryption, a screening step before a model's output is trusted, and the operating evidence we collect weekly. If it's drawn, it runs, with one exception: Atria, which is marked as in development.

reflection elaborative encoding crystallization memory consolidation SESS. SESSION A conversation, captured as it happens EPIS. EPISODE episodic framework, Tulving A topic-bounded unit of experience ECHO ECHO episodic memory trace Contextually bound. Unconsolidated echoes go dormant, never deleted insight what was learned decision what was chosen, and why reference facts worth keeping at hand preference how you like things done ··· more echo types E ENGRAM long-term memory substrate gathered into convergence zones Retrieved by context, never front-to-back. Only what bears on the task at hand. ecphory recall by cue, Tulving The next conversation is the cue; only what bears on it is recalled
SessionA conversation with any connected AI tool, recorded as it happens. Nothing is summarized yet; this is the raw experience.
EpisodeThe session is cut into episodes, each about one thing. Tulving's episodic framework: what happened, when, in what context.
EchoReflection turns an episode into echoes: typed traces such as a decision, an insight, a reference, or a preference. Some stay latent until something later confirms them.
EngramEchoes that keep being confirmed crystallize into engrams, the long-term substrate. Semon's word, Tonegawa's lab work.
Convergence zoneEngrams mass into convergence zones by meaning, so a task lands in a neighborhood of knowledge rather than a flat list.
EcphoryTulving's term for recall triggered by a cue. When a new conversation begins, its context is the cue: the engrams and echoes with bearing on it are activated and injected, and the rest stay dormant. The archive can grow without the cost of recall growing with it.

Sources. Tulving, E. (1972), "Episodic and semantic memory," in Organization of Memory · Müller, G. E. and Pilzecker, A. (1900), Experimentelle Beiträge zur Lehre vom Gedächtnis · Semon, R. (1904), Die Mneme · Tonegawa lab, MIT, engram-cell studies. The measured figures on this site link to the product's benchmarks page. What is drawn is what the stages are; the thresholds and rules that decide what consolidates and what is trusted are not, by design.

Retrieval

A temporal knowledge graph, read by cue.

Memories are stored in a temporal knowledge graph. Every entry keeps its time and its links to what was said with it. Retrieval starts from the prompt in front of you, which is the cue; the graph is read by that cue, and what comes back is what's connected to it. Warm, that read takes 0.2 to 0.4 seconds. Over time, memories are distilled; a distilled memory carries up to 27 times its stored size.

The Anamnesis dashboard
The dashboard shows counts and a live event stream; memory contents are never displayed here.

Atria In development

Accountability

Atria, in development.

Atria is in development. It's the second product, and we're not describing it until there's a measurement to describe. The rule on this site is that a sentence about a product says what a buyer receives. For Atria we can't say that yet, so the entry stays short. When it can, it'll carry numbers the way Anamnesis does.

Accountability.

Your account holds its own keys. You can see every memory, search all of it, and delete any of it. Each recall leaves a receipt in the prompt, so you know when memory was used and how much: "[anamnesis] recalled 3 memories for this prompt — context you didn't have to re-explain." SOC 2 isn't done; weekly operating evidence has been committed since August 2026, and the audit starts when it's funded. The firm is accountable, at hello@smtry.ai.

Need it inside your own systems? SMTRY Applied → · Questions first? Write to us →