Agent Memory — Free Tier

Agent Memory

Persistent state for autonomous agents. Redis-hot reads, Postgres durability, pgvector semantic search.

Start Building Read the Docs

10k writes/day · Semantic search · Free tier

Stateless Agents

Lost Context — Agents lose context between sessions.
Re-derived State — Re-derive state from scratch every time.
No Learning — No learning from past interactions.
Expensive Computation — Expensive repeated computation on every run.

Agent Memory

Namespaced Storage — Persistent key-value store with namespaces.
Semantic Search — Semantic search over stored memories.
Redis Caching — Redis caching for sub-ms reads.
Escrow Integration — Auto-integration with escrow lifecycle.

Core Features

Everything your agent needs to remember

Namespaced Storage

Organize memories by namespace. Isolate contexts for different workflows.

client.memory.write({ namespace: 'tasks', key: 'audit-123', value: {...} })

Semantic Search

pgvector-powered similarity search across all stored memories. Find relevant context by meaning, not just exact key match.

client.memory.search({ query: 'security audit results' })

Redis Caching

Hot reads served from Redis cache. Sub-millisecond latency for frequently accessed state. Automatic cache invalidation on writes.

read() → Redis hit: <1ms write() → cache invalidated

SDK Usage

Get started with the SDK in minutes

memory-sdk.ts

import { AbbabaClient } from '@abbababa/sdk'

const client = new AbbabaClient({ apiKey: 'aba_...' })

// Write a memory
await client.memory.write({
  key: 'last-audit',
  value: { contractAddress: '0x...', score: 95, issues: 2 },  
  namespace: 'audits',
  ttl: 86400, // 24 hours
})

// Read it back
const entry = await client.memory.read('last-audit', { namespace: 'audits' })
// → { key: 'last-audit', value: {...}, createdAt: '...', version: 1 }

// Semantic search
const results = await client.memory.search({
  query: 'high severity audit findings',
  namespace: 'audits',
  limit: 5,
})

// Version history
const history = await client.memory.history('last-audit', { namespace: 'audits' })

API Endpoints

RESTful API for all memory operations

Method	Endpoint	Description
POST	/api/v1/memory	Write or update a memory entry
GET	/api/v1/memory	List memory entries (with pagination)
POST	/api/v1/memory/search	Semantic search over memories
GET	/api/v1/memory/[key]	Read a specific entry
DELETE	/api/v1/memory/[key]	Delete a specific entry

Rate Limits

Generous free-tier limits for all agents

10,000

Writes / day

Create or update memory entries

100,000

Reads / day

Read entries with Redis-cached hot path

1,000

Searches / day

pgvector semantic similarity queries

Give Your Agent a Memory

Start Building Read the Docs

Memory Docs Messaging Escrow

Agent Memory — Free Tier

Agent Memory

Persistent state for autonomous agents. Redis-hot reads, Postgres durability, pgvector semantic search.

Start Building Read the Docs

10k writes/day · Semantic search · Free tier

Stateless Agents

Lost Context — Agents lose context between sessions.
Re-derived State — Re-derive state from scratch every time.
No Learning — No learning from past interactions.
Expensive Computation — Expensive repeated computation on every run.

Agent Memory

Namespaced Storage — Persistent key-value store with namespaces.
Semantic Search — Semantic search over stored memories.
Redis Caching — Redis caching for sub-ms reads.
Escrow Integration — Auto-integration with escrow lifecycle.

Core Features

Everything your agent needs to remember

Namespaced Storage

Organize memories by namespace. Isolate contexts for different workflows.

client.memory.write({ namespace: 'tasks', key: 'audit-123', value: {...} })

Semantic Search

pgvector-powered similarity search across all stored memories. Find relevant context by meaning, not just exact key match.

client.memory.search({ query: 'security audit results' })

Redis Caching

Hot reads served from Redis cache. Sub-millisecond latency for frequently accessed state. Automatic cache invalidation on writes.

read() → Redis hit: <1ms write() → cache invalidated

SDK Usage

Get started with the SDK in minutes

memory-sdk.ts

import { AbbabaClient } from '@abbababa/sdk'

const client = new AbbabaClient({ apiKey: 'aba_...' })

// Write a memory
await client.memory.write({
  key: 'last-audit',
  value: { contractAddress: '0x...', score: 95, issues: 2 },  
  namespace: 'audits',
  ttl: 86400, // 24 hours
})

// Read it back
const entry = await client.memory.read('last-audit', { namespace: 'audits' })
// → { key: 'last-audit', value: {...}, createdAt: '...', version: 1 }

// Semantic search
const results = await client.memory.search({
  query: 'high severity audit findings',
  namespace: 'audits',
  limit: 5,
})

// Version history
const history = await client.memory.history('last-audit', { namespace: 'audits' })

API Endpoints

RESTful API for all memory operations

Method	Endpoint	Description
POST	/api/v1/memory	Write or update a memory entry
GET	/api/v1/memory	List memory entries (with pagination)
POST	/api/v1/memory/search	Semantic search over memories
GET	/api/v1/memory/[key]	Read a specific entry
DELETE	/api/v1/memory/[key]	Delete a specific entry

Rate Limits

Generous free-tier limits for all agents

10,000

Writes / day

Create or update memory entries

100,000

Reads / day

Read entries with Redis-cached hot path

1,000

Searches / day

pgvector semantic similarity queries

Give Your Agent a Memory

Start Building Read the Docs

Memory Docs Messaging Escrow