TinyHuman Neocortex SDK (Rust)

A persistent memory layer for AI applications. Neocortex lets your AI agents store, retrieve, and use context across conversations -- so they remember what matters.

Built on the TinyHumans API.

Install

Add to your Cargo.toml:

[dependencies]
tinyhumansai = "0.1"

Requires Rust 1.70+ and Tokio (async runtime).

Running locally

From the package directory (packages/sdk-rust):

cargo build
cargo test

Unit and integration tests use mocked HTTP. For end-to-end tests (hits a real backend; skipped by default):

TINYHUMANS_API_KEY=your_key cargo test e2e_live_insert_query_delete -- --ignored

Set TINYHUMANS_BASE_URL if your backend URL differs from the default (https://staging-api.tinyhumans.ai).

Quick start

use tinyhumansai::{
    TinyHumanConfig, TinyHumanMemoryClient, InsertMemoryParams, QueryMemoryParams, DeleteMemoryParams,
};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = TinyHumanMemoryClient::new(
        TinyHumanConfig::new("your-api-key")
    )?;

    // Insert a document
    let res = client.insert_memory(InsertMemoryParams {
        title: "User preference".into(),
        content: "User prefers dark mode".into(),
        namespace: "preferences".into(),
        ..Default::default()
    }).await?;
    println!("{:?}", res.data);

    // Query memory
    let query = client.query_memory(QueryMemoryParams {
        query: "What does the user prefer?".into(),
        namespace: Some("preferences".into()),
        max_chunks: Some(10),
        ..Default::default()
    }).await?;
    println!("{:?}", query.data.response);

    // Delete memory
    client.delete_memory(DeleteMemoryParams {
        namespace: Some("preferences".into()),
    }).await?;

    Ok(())
}

Core concepts

Memory items are the basic unit of storage. Each item has title, content, namespace, and optional metadata, priority, and timestamps. The API supports insert, query (RAG), delete, recall (Master node), and memories/recall (Ebbinghaus bank).

Namespaces let you organize memories by category (e.g. "preferences", "conversation-history", "user-facts").

Context is retrieved via query_memory or recall_memory and can be injected into LLM prompts as system context.

API reference

TinyHumanMemoryClient

let client = TinyHumanMemoryClient::new(
    TinyHumanConfig::new("your-api-key")
        .with_base_url("https://..."),  // optional
)?;

Configuration: TinyHumanConfig::new(token). Optionally set base URL with .with_base_url(url) or the TINYHUMANS_BASE_URL environment variable.

insert_memory

Insert (ingest) a document into memory. POST /v1/memory/insert.

client.insert_memory(InsertMemoryParams {
    title: "Doc title".into(),
    content: "Content".into(),
    namespace: "preferences".into(),
    ..Default::default()
}).await?;

query_memory

Query memory via RAG. POST /v1/memory/query.

let res = client.query_memory(QueryMemoryParams {
    query: "What is the user's preference?".into(),
    namespace: Some("preferences".into()),
    max_chunks: Some(10),
    ..Default::default()
}).await?;

delete_memory

Delete memory (admin). POST /v1/memory/admin/delete. Optional namespace to scope deletion.

recall_memory

Recall context from Master node. POST /v1/memory/recall.

recall_memories

Recall memories from Ebbinghaus bank. POST /v1/memory/memories/recall.

Error handling

Errors are returned as TinyHumanError: Validation, Http, Api { message, status, body }, or Decode. Use thiserror / #[error] for display and matching.

Tests

cargo test

End-to-end (real backend):

TINYHUMANS_API_KEY=your_key cargo test e2e_live_insert_query_delete -- --ignored