Memory

Honi provides four tiers of memory, each backed by a different Cloudflare primitive. Use one, two, or all four depending on your agent's needs.

Overview

Tier	Backing	Persistence	Use Case
Working	Durable Object	Session-scoped	Conversation history, scratch state
Episodic	D1 (SQLite)	Permanent	Past conversations, user facts, long-term context
Semantic	Vectorize + Workers AI	Permanent	Similarity search over knowledge, RAG
Graph	edgraph (Durable Object)	Permanent	Entity relationships, knowledge graphs, structural recall

Working Memory (Durable Object)

Working memory is always enabled — no configuration needed. It stores the current conversation messages in the Durable Object's transactional storage.

TypeScript

1// Working memory is automatic — just create an agent
2export const agent = createAgent({
3  name: 'my-agent',
4  model: 'claude-sonnet-4-20250514'
5})

Each session gets its own isolated working memory within the Durable Object. Messages are persisted across requests within a session and cleared on DELETE /chat.

Episodic Memory (D1)

Episodic memory stores structured facts and past conversation summaries in a D1 database. This allows agents to recall information from previous sessions.

TypeScript

1export const agent = createAgent({
name: 'my-agent',
model: 'claude-sonnet-4-20250514',
memory: {
  episodic: { binding: 'MEMORY_DB' }
}
7})

Semantic Memory (Vectorize + AI)

Semantic memory enables similarity-based retrieval over a knowledge base. Honi uses Cloudflare Vectorize for vector storage and Workers AI for embedding generation.

TypeScript

1export const agent = createAgent({
name: 'my-agent',
model: 'claude-sonnet-4-20250514',
memory: {
  episodic: { binding: 'MEMORY_DB' },
  semantic: { vectorize: 'VECTORIZE', ai: 'AI' }
}
8})

Or use the shorthand 'tiered' to enable all three tiers with default bindings:

TypeScript

1memory: 'tiered'  // equivalent to all three tiers with default bindings

Full wrangler.toml

Here's a complete wrangler.toml with all three memory tiers configured:

TOML

 1name = "my-agent"
 2main = "src/index.ts"
 3compatibility_date = "2024-12-01"
 4
 5# Working memory — Durable Object
 6[[durable_objects.bindings]]
 7name = "AGENT"
 8class_name = "AgentDO"
 9
10[[migrations]]
11tag = "v1"
12new_classes = ["AgentDO"]
13
14# Episodic memory — D1
15[[d1_databases]]
16binding = "MEMORY_DB"
17database_name = "agent-memory"
18database_id = "your-database-id"
19
20# Semantic memory — Vectorize + Workers AI
21[[vectorize]]
22binding = "VECTORIZE"
23index_name = "agent-knowledge"
24
25[ai]
26binding = "AI"

Graph Memory →Workflows →