whitepaper: fix server description (two parts)

This commit is contained in:
Ollie Agent 2026-08-04 21:14:40 +02:00
parent 7899ef3cb1
commit 9522579670
1 changed files with 1 additions and 1 deletions

View File

@ -1,3 +1,3 @@
# Ollie Technical Whitepaper
Ollie is an AI agent runtime that exposes sessions, agents, prompts, tools, and state as files in a 9P2000 filesystem. The server is a single Go binary (`olliesrv`); the client library is `cat`, `echo`, and `tail`. Tool scripts, skill files, and prompt templates live on the physical filesystem and are loaded dynamically — nothing is compiled into the binary beyond the agent loop, the LLM dispatcher, and the sandboxed tool runner. Frontends (Emacs, Qt6, acme, shell scripts) all talk to the same 9P namespace; the server does not distinguish between them. The architecture is deliberately minimal: a coordination surface rather than a framework, inspired by Plan 9's principle that resources should be files and programs should compose through the filesystem rather than through APIs.
Ollie is an AI agent runtime that exposes sessions, agents, prompts, tools, and state as files in a 9P2000 filesystem. The server has two parts: `olliesrv`, the 9P daemon that hosts the agent sessions, and `ollie-9p`, the client that frontends use to mount the namespace. Tool scripts, skill files, and prompt templates live on the physical filesystem and are loaded dynamically — nothing is compiled into the binary beyond the agent loop, the LLM dispatcher, and the sandboxed tool runner. Frontends (Emacs, Qt6, acme, shell scripts) all talk to the same 9P namespace; the server does not distinguish between them. The architecture is deliberately minimal: a coordination surface rather than a framework, inspired by Plan 9's principle that resources should be files and programs should compose through the filesystem rather than through APIs.