smaramwbc/statewave

Open-source memory runtime for AI agents — reproducible, provenance-tagged context bundles instead of query-time retrieval. Apache-2.0, self-hosted on Postgres + pgvector, Python + TypeScript SDKs.

What it solves

Statewave provides a durable, structured memory runtime for AI agents to prevent the loss of context between sessions. It replaces fragile, unstructured prompt-stuffing or simple vector database retrieval with a deterministic system that ensures agents remember preferences, decisions, and history across different conversations.

How it works

Statewave operates as infrastructure that runs alongside an application using a "compile-then-use" approach:

  1. Ingest: Raw events (episodes) are recorded in an append-only format.
  2. Compile: These episodes are processed into typed, summarized memories with confidence scores and provenance.
  3. Retrieve: The system assembles ranked, token-bounded context bundles for prompts on demand.
  4. Govern: It allows for the inspection of subject timelines and the deletion of data by subject.

This process is idempotent, meaning the same query against the same subject at the same point in time always produces the same result, eliminating the sampling noise common in query-time retrieval.

Who it’s for

Developers building AI agents (such as customer support or long-running coding agents) who need their systems to remember user-specific data and architectural decisions across multiple sessions without being locked into a specific chatbot framework.

Highlights

  • Provenance: Every memory can be traced back to its original source episode via content-hashed receipts.
  • Deterministic Context: Provides token-bounded bundles that are consistent per subject and task.
  • Pluggable Compilers: Supports both local heuristic compilers (regex) and LLM-based compilers via LiteLLM.
  • Self-Hosted: Runs on Postgres and pgvector, ensuring the user owns the storage and avoiding vendor lock-in.
  • Multi-tenant Support: Includes query-scoped data isolation and region pinning for data residency.
  • Framework Neutral: Integrates via REST API or SDKs with any AI stack, including LangChain, CrewAI, and AutoGen.

Related

  • Project
    Atmosphere/atmosphere

    Atmosphere is a JVM‑based framework that lets you write AI agents as annotated Java classes. It provides real‑time streaming (WebSocket, SSE, gRPC, optional HTTP/3), durable memory, policy‑driven governance, human‑in‑the‑loop approval, and adapters for many LLM providers. Agents can be exposed to browsers, Slack, Discord, MCP, A2A, etc., without rewriting code.

  • Project
    LeoYeAI/talewell

    Talewell is a git‑backed, Markdown‑based long‑term memory system for AI agents. It stores each memory record as a version‑controlled file, provides deterministic offline recall, enforces provenance, and exposes a portable AMP protocol so any agent platform (OpenClaw, Claude Code, Codex, etc.) can use the same memory repo via plugins or a simple file convention.

  • Project
    rtk-ai/icm

    ICM is a Rust‑based, SQLite‑backed memory system for AI coding agents. It stores and recalls project‑specific facts without needing LLM calls, supports keyword or embedding‑based search, integrates with many coding assistants, and achieves benchmark accuracy comparable to larger services.

  • Project
    basicmachines-co/basic-memory

    A persistent, two-way memory system for AI assistants that uses shared Markdown files and the Model Context Protocol (MCP) to create a long-term knowledge graph.

  • Project
    yuruotong1/autoMate

    autoMate is a locally hosted service that gives AI chat clients persistent personal storage (notes, files, reminders, long‑term memory) and a toolbox of real‑world actions (shell, browser, SaaS APIs). It talks to clients via an MCP‑over‑HTTP bridge, works with OpenClaw, Claude Desktop, Cursor, Cline, etc., and can also be used through its own web chat, Chrome extension, Docker container, or binary.