Skip to content

Repository files navigation

moka-1

moka = Mixture of Knowledge Agents. An agentic RSS feed reader that starts with a single docker compose up -d and keeps working in the background. Also the dogfooding project for Plecto. Design principles live in docs/tenets/moka-tenets.md, decisions in docs/adr/.

English · 日本語

Getting started

git clone https://github.com/Kaikei-e/moka-1 && cd moka-1
openssl rand -base64 24 | tr -d '/+=' > secrets/postgres_password.txt  # once per clone (see secrets/README.md)
docker compose up -d --wait

The first start downloads the LLM model (~5.2 GB), so --wait takes a few minutes. If the LLM is down, moka still works as a plain RSS reader (fail-soft).

Layout (5 resident services)

Service Tech Role
plecto Plecto (Rust + WASM filters) Edge: TLS termination, HTTP/1·2·3, routing
moka-core (core/) Single Go binary API + resident agent (fetch / enrich / recap)
moka-web (web/) SvelteKit SSR The reading UI
db (db/) PostgreSQL 18 + pgvector Articles, feeds, embeddings, job state — everything
llm llama.cpp server (Vulkan) Local inference

Migrations are applied automatically at startup by the one-shot migrate job (Atlas).

About

moka = Mixture of Knowledge Agent. An agentic RSS feed reader that starts with a single docker compose up -d and keeps working in the background.

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages