BaseMyAI

The private memory database for AI agents.

Local-first · Temporal · Encrypted · Agent-scoped · Hybrid recall

Repository · Documentation · Security · crates.io · PyPI

---

What is BaseMyAI?

BaseMyAI is an embedded Agent Memory Database for AI agents that need memory they can actually trust.

It stores agent memory inside a local encrypted .bmai database and gives agents persistent recall with vector search, temporal validity, memory layers, graph context, adaptive forgetting, and hybrid retrieval.

BaseMyAI is not another hosted vector database.

It is built for private assistants, coding agents, internal copilots, local-first tools, and products that cannot leak user memory to a cloud memory store.

---

Quick start

basemyai setup --fetch

export BASEMYAI_DB_KEY="change-me-use-a-real-secret-key"

basemyai init ./agent.bmai

basemyai remember ./agent.bmai \
  --agent assistant-42 \
  --layer semantic \
  "User is on the Pro plan."

basemyai recall ./agent.bmai \
  --agent assistant-42 \
  "current billing plan" \
  --hybrid

---

Core principles

Principle	Description
Local-first by default	Memory stays on the user's machine. No hosted memory backend by default.
Encrypted .bmai files	Agent memory lives in a local encrypted BaseMyAI container.
Temporal memory	Facts can expire or be invalidated, so stale memories do not silently win recall.
Agent-scoped isolation	Each memory operation is scoped by agent_id to prevent cross-agent leakage.
Multi-layer memory	Short-term, episodic, procedural, and semantic memories are treated differently.
Hybrid recall	Vector search, BM25, graph context, and RRF fusion work together.
Inspectable memory	Developers can inspect, export, import, invalidate, and debug memory behavior.
Rust foundation	One engine powers Rust, Python, Node, CLI, MCP, and REST surfaces.

---

Why BaseMyAI exists

AI agents usually fail memory in one of three ways:

1. They forget everything between sessions.
2. They send sensitive memory to a hosted vector database.
3. They recall outdated facts as if they were still true.

BaseMyAI fixes that by giving agents a dedicated memory database that is local, encrypted, temporal, and purpose-built for agent workflows.

A vector database can store embeddings.

BaseMyAI stores agent memory.

---

Memory model

BaseMyAI separates memory into four layers:

Layer	Purpose	Typical lifetime
short_term	Active working context	Short TTL
episodic	Events, interactions, what happened and when	Time-bounded
procedural	Workflows, steps, habits, learned behaviors	Long-lived
semantic	Facts and knowledge	Valid until invalidated

Every layer supports temporal validity through valid_from and valid_until.

That means an agent can remember that something used to be true without treating it as true forever.

---

Recall system

BaseMyAI recall is multi-signal by design.

Signal	Role
Vector search	Finds semantically similar memories.
BM25 full-text search	Catches exact terms, IDs, names, filenames, and rare keywords.
Graph context	Connects entities and relationships across memories.
Temporal filtering	Removes expired or invalidated facts from active recall.
RRF fusion	Merges ranking signals without brittle manual weights.

This makes recall more reliable than cosine similarity alone.

---

Security and privacy

BaseMyAI is built for developers who need memory they can control.

• encrypted local .bmai files
• no hosted memory backend by default
• no silent model downloads
• explicit setup for model provisioning
• per-agent memory isolation
• local-first operation
• key supplied by the application or environment
• designed for privacy-sensitive agents and internal tools

See the full security policy:

Security Policy

---

Supported environments

macOS    Linux    Windows

BaseMyAI is built in Rust and designed for cross-platform local deployment.

---

Main repository

BaseMyAI/basemyai

The main repository contains:

• Rust workspace
• memory engine
• CLI
• MCP server
• REST sidecar
• Python binding
• Node binding
• architecture decisions
• benchmarks
• security policy
• technical proofs

---

Consumption surfaces

Surface	Status	Use case
Rust crate	Available	Native Rust agents and embedded engines
Python SDK	Available	Python agent builders
Node SDK	In progress / verify package name	JavaScript and TypeScript agents
CLI	Available	Local inspection, scripting, automation
MCP server	Available	Claude Code and agent-tool integrations
REST sidecar	Available	Local HTTP integration for other languages

---

Product direction

BaseMyAI starts with a reliable libSQL-backed .bmai V1, then moves toward a native BaseMyAI engine while preserving the public memory database contract.

The long-term direction is full ownership over:

• storage
• indexing
• vector search
• graph traversal
• memory events
• future sync primitives
• query language possibilities

The public contract stays simple:

a private memory database for AI agents.

The internal engine can evolve without turning BaseMyAI into a wrapper around someone else's database.

---

Contributing

BaseMyAI is being built as serious local AI memory infrastructure.

Useful places to start:

• explore the main repository
• read the architecture decisions
• run the CLI locally
• inspect the benchmark docs
• open issues or discussions
• test SDK and MCP workflows

Open the repository

---

Useful links

• Repository
• Documentation
• Security Policy