docs.rs failed to build rullama-stores-0.12.0
Please check the build logs for more information.
See Builds for ideas on how to fix a failed build, or Metadata for how to configure docs.rs builds.
If you believe this is docs.rs' fault, open an issue.

rullama-stores

Schema + CRUD for the rullama agent framework's opinionated minimum data-store set.

What this crate is

The framework is opinionated about a small set of stores every agent system needs — sessions, conversations, tasks, plans, locks, images, templates, and the hot/warm/cold memory tiers. This crate defines them all as schema + CRUD only, generic over the [rullama_storage::StorageBackend] trait so you can swap backends (LanceDB, in-memory, your own) without touching store code.

It deliberately contains no orchestration, no engines, no pipelines — just rows and the operations that move them in and out. Multi-tier search, promotion / demotion, and offline consolidation live in the separate rullama-memory crate, which depends on the schema types here.

Features

Flag	Default	What it pulls in
`session`	yes	`SessionStore` trait + `InMemorySessionStore` + (with `sqlite`) `SqliteSessionStore`
`task`	yes	`TaskStore` + `AgentStateStore`
`plan`	yes	`PlanStore` + `TemplateStore`
`conversation`	yes	`ConversationStore` (catalog metadata: id, title, model, message count)
`memory`	no	tier schema stores: `MessageStore`, `SummaryStore`, `FactStore`, `MentalModelStore`, `TierMetadataStore` + `tier_types`
`lock`	no	`LockStore` (rusqlite-backed coordination locks)
`image`	no	`ImageStore` with sha256 hashing
`sqlite`	no	rusqlite backend for `SqliteSessionStore` and `LockStore`

default = ["session", "task", "plan", "conversation"] covers the most common agent-runtime needs. Set default-features = false and opt into memory / lock / image only when you need them.

How the three storage layers fit together

rullama-storage    StorageBackend trait, backends (LanceDB, …),
                      embeddings, BM25, file-context primitives.
                      ─── substrate ───
              ▲
              │
rullama-stores     Row schemas + CRUD for the opinionated minimum
                      set (this crate). Built on the substrate.
                      ─── schema ───
              ▲
              │
rullama-memory     TieredMemory orchestration over the tier
                      schema stores; multi-factor adaptive search;
                      offline `dream` consolidation engine.
                      ─── orchestration ───

SessionStore example

use rullama_stores::{InMemorySessionStore, SessionId, SessionStore};
use rullama_core::Message;

let store = InMemorySessionStore::new();
let id = SessionId::from("user-123");
store.save(&id, &[Message::user("hello")]).await?;
let transcript = store.load(&id).await?;

License