OctaMem · Persistent memory for AI systems

Skip to content<br>Sign inStart free

Navigate<br>01Home<br>02Platform<br>03Pricing<br>04Developers<br>05Trust<br>06About<br>07Docs<br>Get started<br>Start freeTalk to sales<br>Already have an account · Sign in

Vol. 01Memory infrastructure<br>Live · status nominal<br>Edition · 2026<br>An infrastructure for everything your agents shouldn’t forget.<br>Memory is<br>all you need.<br>The persistent memory layer for AI agents. Built for the stacks where forgetting isn’t an option.<br>Start freeTalk to sales<br>No card. 2 GB on the free tier.

Works with the stack you already run· OpenAI<br>OpenAI·<br>MCP·<br>Cursor·<br>LangGraph·<br>Vercel AI SDK

Design partnersInitial cohort in healthcare, finance, defense, and legal.<br>Apply to join

§ 01Problem & solution<br>The cost of forgetting.<br>Abstract<br>№ 01<br>Agents without memory fail in two ways that show up on the balance sheet: they burn tokens re-reading context, and they let hard-won institutional knowledge walk out the door. A memory layer answers both.

01The cost<br>Token spend compounds every turn.<br>Without memory, the same context is re-sent on every call. Conversations re-explain themselves, prompts balloon, and you pay frontier-model rates to re-read what the model was already told an hour ago.

The solution<br>Send less. Repeat nothing. Pay less.<br>Less context per call — only what's relevant is retrieved and injected<br>No repetition — facts and decisions persist instead of being re-sent<br>Cheaper models hold their own once the context they receive is sharper

02The leak<br>Institutional knowledge isn't centralised.<br>What your agents and teams learn lives in scattered sessions, local notes, and individual heads. When an employee leaves, it leaves with them. Nothing compounds, and nothing is owned by the organisation.

The solution<br>One memory the whole organisation owns.<br>Organisation-wide intelligence — every agent reads from one shared layer<br>Knowledge stays when people leave — it lives in the memory, not the person<br>Context compounds across teams instead of resetting every session

Memory has to be infrastructure — not a patch.<br>→See how the architecture solves it

§ From failure to system

§ 02The Architecture<br>Memory in motion.<br>Every request passes through the same disciplined cycle. OctaMem doesn’t fire a generic search across one bucket of text. It rebuilds context from three memory types that each serve a distinct purpose, then reassembles them for the model.<br>Read the technical brief

Search cycleAdd cycle<br>Search cycle · in motion<br>Stage 01 / 05<br>01 / Caller<br>App or MCP

02 / Access · quota<br>Security layer

03 / OctaMem agent<br>Retrieval service

04 / Three layers<br>Memory layers

05 / Back to app<br>Unified context

semanticepisodicprocedural01Appor MCPCaller02SecuritylayerAccess · quota03RetrievalserviceOctaMem agent04MemorylayersThree layers05UnifiedcontextBack to app<br>fig. 1 · Search cycle · stage 1 of 5Auto-playing

Search cycleAdd cycle<br>fig. 1 · Search · stage 1 / 5

01Connect through API, SDK, or MCP.<br>An app, assistant, or MCP client opens the request with a memory API key. Same interface for chatbots, copilots, and pipelines.<br>02Validate before memory is touched.<br>Access, quota, and policy run before retrieval. Sensitive fields never reach the model.<br>03Retrieve from all three memory systems.<br>Semantic, episodic, and procedural records are pulled in parallel, each from its own optimized store.<br>04Unify into one context package.<br>Records are reassembled into a single compact context with citations and retention tags. The model only sees what matters.<br>05Return with continuity.<br>Response goes back to the caller. The decision is captured for the next request, scoped and traceable.

§ 03File ingestion<br>Any file. Now memory.<br>Hand OctaMem the document itself. Contracts, decks, spreadsheets, emails, PDFs. We parse, structure, and store it as typed memory your agents can query forever.<br>Not embeddings of a blob. Clauses, parties, obligations.<br>Batch upload5 files<br>Avg pages40<br>Max file30 MB<br>RetentionConfigurable

Drop the file. Memory does the rest.<br>contract-v3.pdfPDF<br>Master Services Agreementparties · term · obligations

contract-v3.pdf<br>Q3-roadmap.pptx<br>support-thread.eml<br>customers.csv<br>Parsed memory record<br>contract-v3.pdf<br>Master Services Agreement,<br>v3 · executed 2026-04-12<br>›parties: Acme Corp, OctaMem Inc.<br>›term: 24 months, auto-renew 12<br>›obligations: 99.9% uptime SLA, 30-day deletion<br>Searchable across the account under previous_context: legal-msas.

01 / category<br>Documents<br>Contracts, briefs, reports, knowledge bases.<br>Supported<br>.pdf<br>.docx<br>.docm<br>.dotx<br>.dotm<br>.odt<br>.rtf<br>.txt

02 / category<br>Spreadsheets<br>Tables, ledgers, datasets — reasoned over rows.<br>Supported<br>.xlsx<br>.xls<br>.csv

03 / category<br>Presentations<br>Decks parsed slide by slide. Factual recall, not pixels.<br>Supported<br>.pptx

04 / category<br>Email & Data<br>Threads, payloads, structured exports.<br>Supported<br>.eml<br>.json

§ From input to inheritance

§ 04Compounding intelligence<br>Intelligence that compounds.<br>Every session without memory is a reset. Every session with memory is an upgrade.

Day...