zeos-memory 0.1.0

Memory infrastructure for AI agents — persistent context, session governance, and three-tier memory

zeos-memory

Your AI agents remember.

Install

pip install zeos-memory

Quick Start

from zeos_memory import ZeosMemory

zm = ZeosMemory(project_id="my-project", agent="claude")
zm.start_session()
zm.snap(delta="Built the auth module")
zm.snap(delta="Added JWT validation")
journal, memory = zm.end(
    summary="Authentication complete",
    delta="All tests passing",
    next_actions="Add refresh token rotation",
)
print(zm.render_memory())

Features

• Three-tier memory — long-term (SOUL), mid-term (journals), short-term (working state)
• Decay curation — entries auto-archive by relevance score (0-6), pinned entries persist
• Session lifecycle — deterministic create/snap/end with continuity digests
• Decision tracking — first-class entities with rationale and alternatives
• Agent-agnostic — same protocol governs Claude, Gemini, Codex, Kimi
• Zero dependencies — core SDK needs only pydantic

Optional Extras

pip install zeos-memory[mcp]       # MCP server for AI agents
pip install zeos-memory[graphiti]   # Temporal knowledge graph backend
pip install zeos-memory[full]       # Everything

MCP Server

Add to your agent's MCP config:

{
  "mcpServers": {
    "zeos-memory": {
      "command": "python",
      "args": ["-m", "zeos_memory"]
    }
  }
}

Tools: zeos_boot, zeos_snap, zeos_recall, zeos_curate, zeos_end.

Architecture

Tier	Content	Persistence
Long-Term	SOUL identity, MEMORY.md	Permanent, rolling synopsis
Mid-Term	Session journals, blueprints	Session-scoped, summarized
Short-Term	Working state	Ephemeral, checkpointed

Memory entries carry decay scores. Each session decrements non-pinned scores. Over-budget entries auto-archive. The system curates itself.

Protocol

zeos is a governance and persistence protocol for AI agents. It runs IN context, not ON hardware. See the full Protocol Specification.

License

Apache-2.0