Memory and vectors
Per-user memory, and vector indexes to search your own data.
AxAgent has two ways to keep information: memory for facts about each user, and vectors for searching your own data.
Coming Q4 2026
AxAgent launches in Q4 2026. The API below may change before launch.
| Feature | What it stores | Scope |
|---|---|---|
| Memory | Facts about a user | One agent, one user |
| Vectors | Your documents and data | An index you create |
Enable the modules
const agent = await axerity.agents.create({
model: 'anthropic/claude-sonnet-5.5',
modules: {
memory: true,
vectors: { indexes: ['help-center'] },
},
})Memory
Memory is kept per user. When you build an agent for your own customers, each customer gets their own memory. See sessions.
Add a memory
await axerity.memories.add({
agent: agent.id,
user: customer.id,
text: 'Prefers email over phone. Timezone is CET.',
})Search memories
const { data } = await axerity.memories.search({
agent: agent.id,
user: customer.id,
query: 'How should we contact them?',
})Vectors
Vectors let you vectorize your data and search it by meaning. Create an index, add records to it, then query it.
Create an index
const index = await axerity.vectors.indexes.create({
name: 'help-center',
})Add records
Each record has an id and the text to vectorize.
await axerity.vectors.upsert(
index.id,
articles.map((article) => ({ id: article.slug, text: article.body })),
)Query the index
const { matches } = await axerity.vectors.query(index.id, {
text: 'reset my password',
topK: 5,
})List an index in the agent's vectors module to give the agent access to it.