Harness

The harness is the entry point to db0. It scopes all memory, state, and log operations to a specific agent, session, and user.

Creating a harness

import { db0 } from "@db0-ai/core"

const harness = db0.harness({
  agentId: "my-agent",
  sessionId: "session-1",
  userId: "user-123",
  backend,
  profile: PROFILE_AGENT_CONTEXT,   // optional
  embeddingFn,                       // optional
  batchEmbeddingFn,                  // optional
  extraction,                        // optional
  summarizeFn,                       // optional
})

Configuration

Field	Type	Required	Description
`agentId`	`string`	yes	Identifies this agent
`sessionId`	`string`	yes	Identifies the current session
`userId`	`string`	no	Identifies the user — enables user-scoped memory
`backend`	`Db0Backend`	yes	Storage backend (SQLite or PostgreSQL)
`profile`	`Db0Profile`	no	Tuning profile for scoring, extraction, retrieval
`embeddingFn`	`(text) => Promise<number[]>`	no	Custom embedding function
`batchEmbeddingFn`	`(texts) => Promise<number[][]>`	no	Batch embedding for bulk operations
`extraction`	`object`	no	Extraction strategy config
`summarizeFn`	`(text) => Promise<string>`	no	Custom summarization function

Methods

Method	Returns	Description
`memory()`	`Memory`	Access the memory component
`state()`	`State`	Access the state component
`context()`	`Context`	Access the context component
`log()`	`Log`	Access the log component
`spawn(config)`	`Harness`	Create a child harness
`close()`	`Promise<void>`	Close the backend (root only)

Embedding migration

When you switch embedding providers, existing vectors become incompatible. db0 tracks the active embedding provider and can re-embed all memories:

// check current status
const status = harness.embeddingStatus("gemini-001")

// re-embed all active memories with a new provider
await harness.migrateEmbeddings(
  newEmbeddingFn,
  newBatchEmbeddingFn,
  "gemini-001"
)