Skip to content

discoverability.md

Latest commit

 

History

History
119 lines (106 loc) · 3.98 KB

discoverability.md

File metadata and controls

119 lines (106 loc) · 3.98 KB
Raw

Discoverability Metadata

This document captures the metadata and phrasing that best describes the current repository for GitHub, search, and agent-driven discovery.

Recommended GitHub description

Persistent local MCP memory for AI agents: scoped continuity packs, writable notes, closure checks, and token-aware context reuse for Codex, Claude Code, Gemini CLI, Qwen Code, and other MCP-compatible clients.

Recommended GitHub topics

  • codex
  • claude-code
  • gemini-cli
  • mcp
  • model-context-protocol
  • sqlite
  • agent-memory
  • mcp-memory
  • ai-agents
  • ai-agent-memory
  • local-first
  • token-efficiency
  • context-compression
  • agent-workflows
  • deepseek
  • ollama
  • kimi
  • fastapi
  • developer-tools
  • python

Recommended release framing

1.0.2 is the current public maintenance release for the stable 1.0.x local-first MCP core. The 1.0.x line is a stable local-first MCP continuity layer for MCP-compatible AI agents built on SQLite, scoped continuity packs, deterministic closure control, recent-change deltas, scope-guard retrieval, operational-state carry-forward, provenance-aware auditing, health diagnostics, project snapshots, read-only runtime mode, compact MCP responses, pack-hash reuse, lazy SQLite initialization, runtime heartbeat diagnostics, optional telemetry, an optional local daemon/stdio bridge, MCP retrieval, and a local inspection UI.

v1.0.2 also keeps the identity/scope hardening current, while v1.0.1 hardens daemon/stdio transport behavior, documents client response-mode differences, keeps object-root MCP structuredContent for list results, and adds release hygiene gates. 1.0.0 remains the reproducible verification baseline for the original token-reduction fixtures.

codex-agent-mem can also be documented as an optional companion to clean-process-ended (GitHub) v0.7.2. The projects are independent: codex-agent-mem preserves continuity and closure state, while clean-process-ended provides dry-run process-hygiene evidence and compact janitor receipts.

Search phrases this repo should naturally support

  • Codex memory
  • Codex CLI memory
  • Codex Desktop memory
  • Claude Code memory
  • Google Gemini CLI memory
  • Claude Code MCP memory
  • Google Gemini CLI MCP memory
  • Qwen Code MCP memory
  • local MCP memory for MCP-compatible AI agents
  • MCP memory
  • MCP memory layer
  • MCP context pack
  • structuredContent MCP memory
  • local SQLite MCP memory
  • multi-runtime MCP memory
  • local-first MCP continuity
  • agent continuity memory
  • local-first agent memory
  • SQLite memory for Codex
  • Codex notify capture
  • Codex MCP retrieval
  • auditable agent memory
  • memory provenance for Codex
  • Codex memory health
  • Codex memory snapshots
  • governed memory for Codex
  • policy-based agent memory
  • selective memory inheritance
  • repairable agent memory
  • context compression for agents
  • token-saving context reuse for MCP-compatible AI agents
  • token efficiency for AI agents
  • token economy for agent workflows
  • local memory layer for agent workflows
  • persistent local memory for MCP clients
  • MCP memory for Codex
  • Claude Code memory MCP
  • Gemini CLI memory MCP
  • DeepSeek model workflow through MCP memory
  • Qwen Code MCP memory
  • Ollama model workflow with MCP memory
  • Kimi Code MCP memory
  • codex-agent-mem clean-process-ended
  • continuity plus process hygiene for AI agents

Rules for future docs

  • Prefer explicit, real keywords over abstract marketing language
  • Repeat important terms naturally:
    • MCP
    • SQLite
    • local-first
    • agent memory
    • continuity packs
    • token efficiency
    • token-saving context reuse
    • validated MCP clients such as Codex, Claude Code, and Google Gemini CLI
  • Do not document deferred areas as implemented
  • Keep the first section of public docs specific and concrete
  • When using model-route keywords such as DeepSeek or Ollama, state the validated MCP client path and do not imply a native Ollama or DeepSeek MCP adapter in v1.0.x.