GENYS · Memory
Memory Capsules
Capsules are GENYS’s long‑memory primitives—durable, scoped facts with provenance. They survive sessions, can be audited, and power retrieval that composes the right context at the right time.
GENYS currently powers DesignAdvertise.ai—shipping campaigns that learn, adapt, and scale. This page shows the memory layer those campaigns run on.
Simple
- Stop repeating yourself. Capsules remember users, preferences, and decisions across sessions.
- Your voice stays your voice. Outputs align with approved tone and facts.
- Cheaper & faster. Reuse stored truths instead of re-sending huge prompts.
- You own it. Set retention, export on demand, and forget what you don’t want kept.
Technical
- Capsule Graph (episodic + semantic): append‑only facts, summaries, and decisions with provenance & scope.
- RAG++ orchestration: multi‑store retrieval with recency/authority/policy weighting; compact context synthesis.
- Deterministic guardrails: ToneGuard checks candidates against Brand Vector Hash + capsule truth before output.
- Telemetry loop: drift, token burn, latency, and success events adapt retrieval weights and cache policy.
Schema at a glance
A capsule is an append‑only record. Writes are signed; reads are auditable.
Field | Type | Description |
---|---|---|
id | string | Snowflake/ULID identifier |
subject | string | Entity the capsule is about (user/account/asset) |
data | object | Normalized facts, summaries, decisions, tone |
provenance | object | source, author, method, confidence |
scope | object | tenantId, app, user (RBAC applies) |
retention | object | ttlDays, legalHold, redactionNotes |
tags | string[] | Query aides (e.g., campaign:Q3, tone:bold-power) |
createdAt | datetime | Write timestamp |
version | int | Monotonic append version |
signature | string | Signed hash of write payload |
// JSON shape (example) { "id": "cap_01HY...", "subject": "user_8321", "data": { "plan": "Pro", "tone": "bold-power", "last_campaign": "Q3 Launch" }, "provenance": { "source": "app", "actor": "system", "confidence": 0.98 }, "scope": { "tenantId": "t_42", "app": "designadvertise", "user": "user_8321" }, "retention": { "ttlDays": 365, "legalHold": false }, "tags": ["campaign:Q3","industry:legal"], "createdAt": "2025-08-12T10:00:00Z", "version": 7, "signature": "hash_7edbd989..." }
Quickstart
Store → Retrieve → Guard → Generate → Telemetry → Improve.
// 1) Store durable facts (capsule) await genys.capsules.create({ subject: userId, data: { plan: 'Pro', tone: 'bold-power', last_campaign: 'Q3 Launch' }, scope: { tenantId, app: 'designadvertise' }, retention: { ttlDays: 365 } }); // 2) Generate with retrieval + guardrails const reply = await genys.generate({ subject: userId, input: 'Write a follow-up for Q3 Launch.', retrieve: { k: 8, from: ['capsules','events','docs'] }, guards: ['toneguard:brand_voice_prime','facts:capsule_consistency'] }); // 3) Stream telemetry for learning await genys.telemetry.track({ event: 'followup_sent', subject: userId, metrics: { clicked: true, timeToSend: 12 } });
Governance & Trust
FAQ
How are capsules different from just a bigger context window?
Windows forget and get expensive. Capsules persist distilled facts with provenance and scope, then retrieval selects only what matters.
Do capsules memorize private things forever?
No. Retention is explicit. You set TTLs and redaction rules; every write is signed and auditable.
Can responses drift off brand?
ToneGuard compares candidate text to your Brand Vector Hash and capsule truth before output. If it drifts, we correct or block.