AKADATA LIMITED

How AKAMAN Works

How AKAMAN retrieves bounded documentation fragments from native local sources for coding agents.

AKAMAN

A small Linux C utility that retrieves the compact documentation fragment an AI agent needs from the host where it is running.

Download AKAMAN v0.1.0 source →

How it works

Local documentation

Native man pages, package docs, and headers

AKAMAN

Bounded lookup and focused extraction

AI model

Receives only the useful fragment

AI coding agent
      │
      │ MCP request
      ▼
AKAMAN
      │
      ├── native man
      ├── /usr/share/doc
      └── system headers
      │
      ▼
focused extraction
      │
      ▼
bounded authoritative fragment
      │
      ▼
AI context

AKAMAN has no database, copied documentation corpus, vector store, embeddings, cloud lookup, or required AI-generated summary. It delegates discovery and rendering to the host’s native tools. Internally:

Native discovery — the command name is resolved through the host’s own man (and include roots for headers, /usr/share/doc for docs).
Source selection — a chosen source is honoured; a failed lookup never silently switches to another source.
Matching — the query is matched against page names, headings, options, or symbols.
Whole-line / declaration extraction — results are kept as complete lines or declaration blocks, not partial fragments.
Bounded responses — output is budget-capped (around 400 estimated tokens) so reference material stays small.
Progressive disclosure — a bare query returns a synopsis and a section map, so the agent can ask for more only when needed.
No secondary index — there is no copied or re-rendered documentation authority; the installed system stays authoritative.
Predictable no-match — when a requested source has no match, AKAMAN reports no match rather than substituting a different authority.