Persistent memory for AI agents. Single binary. Local-first. Runs offline.

For AI Agents — Claude, Cursor, GPT, LangChain, AutoGPT, robotic systems, or your custom agents. Give them memory that persists across sessions, learns from experience, and runs entirely on your hardware.

We built this because AI agents forget everything between sessions. They make the same mistakes, ask the same questions, lose context constantly.

Shodh-Memory fixes that. It's a cognitive memory system—Hebbian learning, activation decay, semantic consolidation—packed into a single ~17MB binary that runs offline. Deploy on cloud, edge devices, or air-gapped systems.

Choose your platform:

Real-time activity feed, memory tiers, and detailed inspection

Knowledge graph visualization — entity connections across memories

Projects and todos with GTD workflow — contexts, priorities, due dates

Built-in task management following GTD (Getting Things Done) methodology:

# Add todos with context, projects, and priorities
memory.add_todo("Fix authentication bug", project="Backend", priority="high", contexts=["@computer"])

# List by project or context
todos = memory.list_todos(project="Backend", status=["todo", "in_progress"])

# Complete tasks (auto-creates next occurrence for recurring)
memory.complete_todo("SHO-abc123")

MCP Tools for Claude/Cursor:

• add_todo — Create tasks with projects, contexts, priorities, due dates
• list_todos — Filter by status, project, context, due date
• complete_todo — Mark done, auto-advances recurring tasks
• add_project / list_projects — Organize work into projects

Experiences flow through three tiers based on Cowan's working memory model:

Working Memory ──overflow──▶ Session Memory ──importance──▶ Long-Term Memory
   (100 items)                  (500 MB)                      (RocksDB)

Cognitive Processing:

• Hebbian learning — Co-retrieved memories form stronger connections
• Activation decay — Unused memories fade: A(t) = A₀ · e^(-λt)
• Long-term potentiation — Frequently-used connections become permanent
• Entity extraction — TinyBERT NER identifies people, orgs, locations
• Spreading activation — Queries activate related memories through the graph
• Memory replay — Important memories replay during maintenance (like sleep)

Claude Code (CLI):

claude mcp add shodh-memory -- npx -y @shodh/memory-mcp

Claude Desktop / Cursor config:

{
  "mcpServers": {
    "shodh-memory": {
      "command": "npx",
      "args": ["-y", "@shodh/memory-mcp"],
      "env": {
        "SHODH_API_KEY": "your-api-key"
      }
    }
  }
}

Key MCP Tools:

• remember — Store memories with types (Observation, Decision, Learning, etc.)
• recall — Semantic/associative/hybrid search across memories
• proactive_context — Auto-surface relevant memories for current context
• add_todo / list_todos — GTD task management
• context_summary — Quick overview of recent learnings and decisions

Config file locations:

pip install shodh-memory

from shodh_memory import Memory

memory = Memory(storage_path="./my_data")
memory.remember("User prefers dark mode", memory_type="Decision")
results = memory.recall("user preferences", limit=5)

Full Python documentation →

[dependencies]
shodh-memory = "0.1"

use shodh_memory::{MemorySystem, MemoryConfig};

let memory = MemorySystem::new(MemoryConfig::default())?;
memory.remember("user-1", "User prefers dark mode", MemoryType::Decision, vec![])?;
let results = memory.recall("user-1", "user preferences", 5)?;

Full Rust documentation →

Have an implementation? Open a discussion to get it listed.

[1] Cowan, N. (2010). The Magical Mystery Four: How is Working Memory Capacity Limited, and Why? Current Directions in Psychological Science.

[2] Magee, J.C., & Grienberger, C. (2020). Synaptic Plasticity Forms and Functions. Annual Review of Neuroscience.

[3] Subramanya, S.J., et al. (2019). DiskANN: Fast Accurate Billion-point Nearest Neighbor Search. NeurIPS 2019.

Apache 2.0

MCP Registry · PyPI · npm · crates.io · Docs