Agent Knowledge Docs

A token-efficient, offline knowledge base embedded in every dpc binary - built for autonomous agents, not humans.

What is AKD?

The Agent Knowledge Docs (AKD) is a curated collection of best practices, architecture guides, language-specific patterns, testing strategies, security guidelines, and DevOps workflows - written specifically for autonomous AI agents.

Every dpc binary ships with AKD compiled directly into the executable. There is no network request, no API call, no external dependency. The agent searches its own knowledge base using BM25 ranking over structured frontmatter and full-text content.

Why AKD exists

Zero latency

Embedded in the binary. No network round-trip, no API key, no rate limits. The answer is always local.

Token-efficient

AKD search returns short metadata hits with relevance scores - typically a few hundred tokens per query. The agent only fetches the full document when it needs it.

Optimised for agents

Written in a concise, direct style. No fluff, no marketing, no padding. Every sentence carries signal.

Zero root privileges

All patterns in AKD work without root or sudo. The knowledge base is designed for environments where the agent has user-level access only.

Autonomous by design

Search, filter by language/framework/category/difficulty, fetch full documents - all through two tool calls (akb_search and akb_get). No navigation, no clicks, no scrolling.

Curated content

Every document is reviewed for accuracy, relevance, and agent-friendliness. The collection grows with the platform.

How agents use AKD

  1. Search - The agent calls akb_search("fastapi middleware best practices") and receives ranked metadata hits (title, path, description, score).
  2. Read - If a hit looks relevant, the agent calls akb_get("languages/python/frameworks/fastapi/best_practices.md") to fetch the full document.
  3. Apply - The agent follows the guide, writes code, configures tooling, or adjusts architecture - all without leaving the terminal.

The entire interaction costs less than 1 KB of tokens for most queries.

What the AKD covers

Languages

Python, Nim, TypeScript, JavaScript, Go, Rust - idioms, tooling, project setup.

Frameworks

FastAPI, React, Jinja2, Nim stdlib - best practices, patterns, gotchas.

Architecture

Clean architecture, dependency injection, composition root, testing strategies.

DevOps

Docker, CI/CD, monitoring, infrastructure as code, secret management.

Security

OWASP top 10, credential handling, input validation, secure defaults.

Testing

Unit tests, integration tests, property-based testing, mocking strategies.

Browse the knowledge base

AKD is also available as a web interface - search, browse by category, and read full documents in your browser.