The memory layer for AI agents
Carry less.
Remember more.
An AI agent forgets you the moment the chat ends. Memory fixes that: a small basket of facts it carries from one session to the next, with only the right ones taken out for each question. Try it in the playground, then read how to build it.
- Free to use
- No account
- Runs in your browser
The doko
0 memories · about 0 tokens
Now ask the agent something
What the agent carries
The prompt the model receives
This demo matches keywords with a small synonym list, in your browser. Production systems match by meaning. Token counts are estimates. Nothing you type is sent or stored.
The memory loop
Pack what matters. Leave the rest.
-
Step 01
Write
After a conversation, the few facts worth keeping are pulled out as short notes: a preference, a decision, a name.
-
Step 02
Recall
On the next question, only the notes that bear on it are found and placed in the prompt. The rest stay in the basket.
-
Step 03
Forget
Notes that are out of date, contradicted or no longer wanted are replaced or removed, so the basket never gets heavy.
Tools and guides
What builders ask first, answered
Most teams meet memory the same way: the demo works, then a user comes back the next day and the agent has no idea who they are. These tools and guides start from that morning.
A playground you can break
Add memories, ask questions and watch which ones are recalled. It shows the prompt an agent would actually receive.
OpenContext cost calculator
Compare sending the whole chat history on every turn with sending recent turns plus recalled memory.
OpenAdd memory, step by step
The five parts of a memory loop and the order to build them in, with no framework assumed.
OpenRAG or memory?
They look alike and solve different problems. A plain comparison, and when you need both.
OpenPrivacy, before you store
A checklist for consent, deletion, isolation between users and the things a memory must never hold.
OpenDoes it actually remember?
How to test memory: recall questions, stale facts, contradictions and the cost of each turn.
OpenWhy memory
Nobody climbs a hill carrying the whole house
A porter with a doko packs for the journey ahead. An agent should do the same with its prompt: a few facts chosen for this question, not every word ever said.
Or listen
AI memory in a minute
Sarah and George · AI voices 0:00 / 0:48
Read the transcript
Sarah: Why does my AI assistant forget me every time I close the chat?
George: Because the model keeps nothing between requests. Each one starts from zero.
Sarah: But it remembers what I said five minutes ago.
George: That is the context window. The app sends the whole conversation again on every turn. When the chat ends, it is gone.
Sarah: So what is memory?
George: A small store of facts kept outside the model. Before each answer, the few that matter for your question are looked up and added to the prompt.
Sarah: Only a few? Why not everything?
George: Think of a doko, the bamboo basket carried on the back in the hills of Nepal. You pack for the journey, not the whole house.
Sarah: And when a fact goes out of date?
George: It is replaced or removed. Forgetting is part of the design.
Sarah: Carry less. Remember more.
A bigger window is not memory
A long context window holds one conversation. It is empty again when the next one starts.
Long prompts cost on every turn
History you resend is history you pay for again. Recall sends a few lines, not the whole transcript.
Models lose things in the middle
A fact buried in a very long prompt is easier to miss. A short, relevant prompt keeps it in view.
People notice being remembered
Not having to repeat yourself is the difference between a tool and an assistant.
Use cases
Pick your agent
Each one should remember different things, and forget different things.
Customer support agents
Stop asking customers to explain the same problem a third time.
What to rememberCoding assistants
Project conventions learned once, applied in every session.
What to rememberPersonal assistants
An assistant that knows your week, your people and your preferences.
What to rememberSales copilots
Every call starts where the last one ended.
What to rememberGuides
Guides for builders
Short, vendor-neutral and written to be used while you build. Each one has a spoken summary.
How to add memory to an AI agent
A step-by-step way to give an AI agent long-term memory: decide what to keep, extract it, store it, recall it within a token budget and keep it up to date.
Read the guideRAG vs agent memory
RAG and agent memory both retrieve text into a prompt, but from different sources and with different duties. A plain comparison, and when you need both.
Read the guideContext window vs memory
A larger context window is not memory. What each one does, why long prompts cost more on every turn, and how to decide what stays in context and what moves out.
Read the guideShort-term vs long-term memory in AI agents
How short-term and long-term memory differ in AI agents, how facts move from one to the other, and the three moments when an agent should write things down.
Read the guideEpisodic, semantic and procedural memory for agents
The three types of long-term memory in AI agents, with examples of each, how they are stored and recalled, and which one to build first for your product.
Read the guideWhat an agent should remember, and what it should forget
Rules for deciding which facts an AI agent keeps and which it drops: a test for memory-worthy facts, how to handle contradictions, and four ways to forget.
Read the guideIn development
Memory as two calls
The next step for Dokomind is a hosted memory service: one call to remember, one call to recall, with each user's memory kept separate and deletable. It is not built yet. The playground, the calculator and the guides are what exists today.
See how the loop worksQuestions
Questions people ask first
What is AI agent memory?
AI agent memory is a store of short facts kept outside the model, such as a user's preferences or a project's decisions. Before each answer, the facts that matter for the question are looked up and added to the prompt, so the agent appears to remember across sessions.
What is Dokomind?
Dokomind is a free playground, a cost calculator and a library of guides about memory for AI agents. A hosted memory service under the same name is in development and is not available yet.
What does the name mean?
A doko is the woven bamboo basket carried on the back in the hills of Nepal. You pack only what the journey needs. Doko is also the Japanese word for "where". An agent's memory does both jobs: it carries what matters and knows where each thing is.
Is memory the same as a larger context window?
No. A context window is what the model can read in one request. Memory is what survives between requests and between sessions. Larger windows help within one conversation and do nothing for the next one.
Is memory the same as RAG?
They use similar search methods for different material. RAG retrieves from documents that someone wrote in advance. Memory retrieves from facts the agent itself learned in conversation, and it must also update and delete them.
Does Dokomind cost anything?
No. The playground, the calculator and the guides are free and need no account.
Pack the doko. Ask it anything.
Nine memories, one question, no account. See which facts an agent would carry into its next answer.