tensory

Embedded, claim-native memory for AI agents. Single SQLite file. Built-in collision detection.

82.2% LoCoMo accuracy | 330+ tests | pyright strict | MIT

[](https://pypi.org/project/tensory/) [](https://pypi.org/project/tensory/) [](LICENSE) [](https://github.com/kryptogrib/tensory/actions/workflows/ci.yml)

Plugin | Benchmarks | API Guide | Architecture | Dashboard

Memory works automatically — no tool descriptions in context, no manual search calls:

claude plugin install --source github kryptogrib/tensory --path plugins/claude-code

On install, Claude Code asks for your API keys. That's it — memory activates on every session.

> Full plugin docs: plugins/claude-code/README.md

pip install tensory

from tensory import Tensory, Claim

store = await Tensory.create("memory.db")
await store.add_claims([Claim(text="EigenLayer has 50 team members", entities=["EigenLayer"])])
results = await store.search("EigenLayer")

Extras: pip install "tensory[mcp]" | "tensory[ui]" | "tensory[all]"

<details> <summary><b>MCP server</b> — for Claude Desktop, Cursor, and other MCP clients</summary>

{
  "mcpServers": {
    "tensory": {
      "command": "uvx",
      "args": ["--from", "tensory[mcp]", "tensory-mcp"],
      "env": {
        "TENSORY_DB": "~/.local/share/tensory/memory.db",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Provides 7 tools: tensory_add, tensory_remember, tensory_search, tensory_timeline, `tensory_sta