[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:okf-agent-memory":3},"\u003Ch1>OKF Agent Memory\u003C\u002Fh1>\n\u003Cblockquote>\n\u003Cp>\u003Cstrong>A Domain-Neutral, Git-Native Persistent Project Memory for AI Agents based on the Open Knowledge Format (OKF) v0.2.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FGoogleCloudPlatform\u002Fknowledge-catalog\u002Fblob\u002Fmain\u002Fokf\u002FSPEC.md\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSpecification-OKF_v0.2-blue.svg\" alt=\"Specification\" \u002F>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fokf-memory\u002Fokf-agent-memory\u002Fblob\u002FHEAD\u002Fpkg\u002Fokf\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTooling-Go_1.26_%7C_Zero_Deps-00ADD8.svg\" alt=\"Tooling\" \u002F>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fokf-memory\u002Fokf-agent-memory\u002Fblob\u002FHEAD\u002Fcmd\u002Fokf\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMCP-Ready-purple.svg\" alt=\"Protocol\" \u002F>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fokf-memory\u002Fokf-agent-memory\u002Fblob\u002FHEAD\u002FLICENSE\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-green.svg\" alt=\"License\" \u002F>\u003C\u002Fa>\u003C\u002Fp>\n\u003Chr \u002F>\n\u003Ch2>🌟 Overview\u003C\u002Fh2>\n\u003Cp>Conversations with AI agents reset when context windows close. Valuable architectural decisions, domain discoveries, and operational facts are lost unless stored persistently.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>OKF Agent Memory\u003C\u002Fstrong> provides a standardized, vendor-neutral memory layer that lives directly in your repository (\u003Ccode>knowledge\u002F\u003C\u002Fcode>) as plain Markdown files with YAML frontmatter. It bridges the gap between unstructured ad-hoc markdown files (\u003Ccode>CLAUDE.md\u003C\u002Fcode>, \u003Ccode>AGENTS.md\u003C\u002Fcode>) and complex, black-box vector databases.\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-mermaid\">flowchart TD\n    L1[\"1. OKF v0.2 Specification<br\u002F>(Normative Markdown & YAML Format)\"]\n    L2[\"2. Agent Memory Convention<br\u002F>(Behavioral Rules: Search, Review, Trust)\"]\n    L3[\"3. Agent Skill<br\u002F>(LLM Prompts & Operational Workflows)\"]\n    L4[\"4. Tooling Layer: Go Library & CLI<br\u002F>(Deterministic Parsing, Validation, Search, MCP)\"]\n    L5[\"5. Project Knowledge Corpus<br\u002F>(knowledge\u002F OKF Bundle)\"]\n\n    L1 --> L2\n    L2 --> L3\n    L3 --> L4\n    L4 --> L5\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Chr \u002F>\n\u003Ch2>⚡ Key Highlights\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Blazing Fast Performance (<300µs Search, ~4ms Graph Validation)\u003C\u002Fstrong>: In-memory BM25 retrieval and bundle validation execute in microseconds without VM spin-up or network roundtrips.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>100% Git-Native & Zero Vendor Lock-in\u003C\u002Fstrong>: Everything is version-controlled plain text. Inspect, audit, and review your agent's memory using standard \u003Ccode>git diff\u003C\u002Fcode> and \u003Ccode>git log\u003C\u002Fcode>. No external database required.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Zero API Costs for Memory Retrieval\u003C\u002Fstrong>: Local lexical BM25 indexing eliminates recurring vector embedding API costs and network roundtrips.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Built on Google OKF v0.2\u003C\u002Fstrong>: Uses the open standard format for agent knowledge with full support for provenance (\u003Ccode>sources\u003C\u002Fcode>), trust tiers (\u003Ccode>generated\u003C\u002Fcode> vs. \u003Ccode>verified\u003C\u002Fcode>), and lifecycle metadata (\u003Ccode>status\u003C\u002Fcode>, \u003Ccode>stale_after\u003C\u002Fcode>).\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Solves Context Bloat & Memory Rot\u003C\u002Fstrong>: Employs \u003Cstrong>Progressive Disclosure\u003C\u002Fstrong> (hierarchical \u003Ccode>index.md\u003C\u002Fcode> files and link graphs) so agents only load the exact concepts they need.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Search-Before-Write Principle\u003C\u002Fstrong>: Mandates querying existing memory before authoring, preventing concept duplication and hallucinated divergence.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Zero-Dependency Go Toolchain\u003C\u002Fstrong>: Single binary with \u003Cstrong>zero external dependencies\u003C\u002Fstrong>, sub-5ms CLI startup time, and a built-in \u003Cstrong>Model Context Protocol (MCP) server\u003C\u002Fstrong> (\u003Ccode>okf mcp\u003C\u002Fcode>).\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Truly Domain-Neutral\u003C\u002Fstrong>: Designed for Software Engineering, Coaching, Scientific Research, Literature Reviews, and Operations.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Chr \u002F>\n\u003Ch2>📊 Performance Benchmarks\u003C\u002Fh2>\n\u003Cp>Built in Go with zero external dependencies, \u003Ccode>okf\u003C\u002Fcode> is engineered for high-frequency agent tool calling loops:\u003C\u002Fp>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth align=\"left\">Benchmark Metric\u003C\u002Fth>\n\u003Cth align=\"left\">Python \u002F Vector DB Runtimes (Mem0, Letta)\u003C\u002Fth>\n\u003Cth align=\"left\">Deno \u002F Node.js Tooling\u003C\u002Fth>\n\u003Cth align=\"left\">\u003Cstrong>OKF Agent Memory (Go)\u003C\u002Fstrong>\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd align=\"left\">\u003Cstrong>Concept Search Latency\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd align=\"left\">150ms – 800ms (Embedding API + Vector DB)\u003C\u002Ftd>\n\u003Ctd align=\"left\">40ms – 120ms\u003C\u002Ftd>\n\u003Ctd align=\"left\">\u003Cstrong>< 300 µs (Microseconds, In-Memory BM25)\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"left\">\u003Cstrong>Full Corpus Parse & Graph Validation\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd align=\"left\">200ms – 1.5s\u003C\u002Ftd>\n\u003Ctd align=\"left\">80ms – 250ms\u003C\u002Ftd>\n\u003Ctd align=\"left\">\u003Cstrong>~4.0 ms (50+ concepts, bidirectional graph)\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"left\">\u003Cstrong>Process Cold-Start Overhead\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd align=\"left\">250ms – 600ms (Python VM boot)\u003C\u002Ftd>\n\u003Ctd align=\"left\">80ms – 180ms (V8 \u002F Deno boot)\u003C\u002Ftd>\n\u003Ctd align=\"left\">\u003Cstrong>< 4 ms (Compiled Single Binary)\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"left\">\u003Cstrong>Retrieval Cost per 1,000 Queries\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd align=\"left\">~$0.10 – $0.50 (Embedding tokens)\u003C\u002Ftd>\n\u003Ctd align=\"left\">$0.00\u003C\u002Ftd>\n\u003Ctd align=\"left\">\u003Cstrong>$0.00 (Zero API cost, fully local)\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"left\">\u003Cstrong>Memory Footprint (RSS)\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd align=\"left\">~120 MB – 350 MB\u003C\u002Ftd>\n\u003Ctd align=\"left\">~60 MB – 140 MB\u003C\u002Ftd>\n\u003Ctd align=\"left\">\u003Cstrong>< 15 MB\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Cblockquote>\n\u003Cp>[!TIP]\n\u003Cstrong>Reproduce Locally with your own LLM\u003C\u002Fstrong>: We provide an automated benchmark runner in pure Go to verify Time-To-First-Token (TTFT) speedups and -\u003C\u002Fp>\n\u003C\u002Fblockquote>\n",1789121864033]