Code-aware memory
Korium has a separate commit-linked symbol index. Ask any supplier to show the supported languages, what leaves the machine, how results identify the indexed commit and what happens when code moves.
Korium / Compared with Mem0, Zep and Letta
What do you need your agents to remember?
Remembering a customer’s preferences is one job. Keeping a team’s decisions, reasons and corrections available across its work, and to every assistant it uses, is another. Here is how Korium, Mem0, Zep and Letta compare, row by row, and what each costs.
On this page
The comparison.
| Feature | Korium | Mem0 | Zep / Graphiti | Letta |
|---|---|---|---|---|
| Works from Claude, Codex, Gemini and any MCP tool | YesOne address, browser sign-in | PartlyMem0 MCP and SDKs | YesContext MCP Server | Ask |
| Memory has types: decision, lesson, correction, handoff | YesEleven types, searchable by type | PartlyYour own categories | PartlyCustom entity and edge types | PartlyLabels you choose |
| Keeps who decided and the source | YesDecider, evidence and source on every memory | Ask | PartlySource episodes | Ask |
| A correction keeps the history | YesReplaces, keeps the old, undo | PartlyAdds only; an expiry date hides | YesDated fact invalidation | PartlyGit history of memory files |
| Old memories fade in rank, never vanish | YesOne setting for the organization | PartlyExpiry date only | Ask | Ask |
| Links people, projects and things | YesFound automatically, correctable | PartlyGraph memory on custom plans | YesEntity and relationship graph | Ask |
| Collective memory: a re-save refreshes, a copy folds in, a replacement retires | YesOn paid plans | Ask | Ask | Ask |
| Meeting transcripts, by speaker | YesWho said what, searchable | Ask | PartlyEpisodes | Ask |
| Search answers at the level asked | YesA list, the exact words, your position, or how two things connect | PartlyFilters and score details | PartlyHybrid and temporal | Ask |
| Index of your own code, with descriptions | YesSymbols per branch, described, never file bodies | Ask | Ask | Ask |
| Import what you already have | YesPrepared memories, in batches | YesRaw saves | YesEpisodes | Ask |
| Skills and hooks for your assistant | YesStarter toolkit | Ask | Ask | Ask |
| Hosted, nothing to run | Yes | YesOr self-host | YesOr open-source Graphiti | Yes |
| Reading is never billed | YesUnlimited reads on every plan | PartlyRetrievals counted | PartlyCredits per request | Ask |
| Free tier | Yes$0 a month, 300 saves a month | YesMonthly add and retrieval limits | Yes10,000 credits a month | Yes$0 a month |
| Paid plans start at | $29 a month. Team $125 a month, 5 included. Enterprise custom. | Custom; usage past the free limits | $125 a month for 50,000 credits | $20 a month; Teams $20 a seat |
Sources for the columns: Korium behavior, Memories Fade and the code index; Mem0 saves and search; Zep / Graphiti; Letta memory and sharing. Prices were read on October 26, 2026 from each supplier’s pricing page, Mem0, Zep and Letta, and from Korium’s own; they change, so check before you buy.
The trade-off in one paragraph. Mem0 and Zep are built to be embedded in an application you write, and Letta to build an agent that keeps its own state. Korium is built to be used, not embedded: your people and the assistants they already have share one memory, with the reason, the decider and the history on every record, and the code index beside it. If you are writing an application that must remember its users, start with Mem0 or Zep. If your team keeps re-explaining its own decisions to each other and to their agents, start with Korium.
Where Korium fits better.
- Your team keeps explaining the same decisions. A new colleague or assistant needs to know what failed, what was decided and why a check exists. Korium keeps that as decisions, lessons, corrections and handoffs, each with the reason, the decider and the source, and every connected assistant reads the same record.
- Several assistants, one memory. Claude, Codex, Gemini and any MCP tool share one account at one address. A memory records its owner and who may read it, and an assistant reads only what its person’s role allows.
- Decisions change. A replacement retires the old memory at once and keeps it as history, with undo. A finding never retires a decision, a rejection never expires, and age only lowers rank, by one setting your admin controls.
- Meetings and code. A saved transcript is searchable by who said what. korium-cli indexes your repositories per branch, symbols only, with written descriptions, and the assistant sees the commit an answer came from.
- You do not want to run anything. There is no SDK to embed and no store to operate. Reading is never billed, and the Free plan needs no card.
Where Mem0, Zep or Letta still make sense.
- Mem0: an application that should remember its users. It takes messages, extracts memories and lets your application retrieve them by user, agent, session and category, with a managed API or the open-source SDK in your own environment. How it saves.
- Zep: the question depends on what was true when. Graphiti builds a graph of entities and relationships from source episodes, with time validity on each fact, as a managed service or the open-source framework. How Graphiti works.
- Letta: you are building the agent itself. The Agent SDK gives an agent persistent memory in Git-backed files it edits, with shared repositories for agents on Letta Cloud. Letta memory.
None of this is a benchmark. A storage model is not proof of a correct answer, and hosted products, open-source editions and older SDKs can behave differently. The next section is the test we would run on any of them, Korium included.
The claims we would make every supplier prove.
A comparison helps you choose what to try. Use a task from your own work to check the behavior, and use the supplier’s security and contract documents to check the commitments.
Corrections, history and conflict handling
Save a decision, change it and ask for both current guidance and history. Inspect the actual replacement link and source; do not accept a persuasive explanation with no traceable record.
Self-improvement
Saving learning or updating an agent’s memory does not prove that a system safely improves its own behavior. Ask what changes, what is measured, who approves it and how a bad change is reversed.
Deployment and tenant isolation
A local SDK, an open-source framework and a managed multi-tenant service are different commitments. Korium’s database enforces tenant separation, but a memory scope is not a private compartment within a team; confirm the boundaries your installation needs.
Security and procurement
Ask for the current audit report, contract, data location, subprocessors and deletion terms that cover the service you would buy. Start with our published security information.
And the memory already inside Claude / OpenAI / Gemini?
That is a separate comparison. Their platform memory stays within their products by design; it is not automatically shared company memory for another platform.
Korium gives authorized connected assistants a common place to save and retrieve knowledge. It does not copy private conversations or install a connector for every tool in the diagram. We have also tested OpenClaw; check each client’s supported connection and permissions.
Read the platform-memory questions →Start with the mistake you do not want to repeat.
Use one real, permitted example: a failed approach, the decision that followed and a later correction. Give each system the same material and ask a second agent to make the next plan.
Can it find the source, apply the current decision and explain the change without you teaching it again? That tells you more than a table full of ticks.