Skip to content
View dialectforge's full-sized avatar
3 followers · 19 following

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: YOLOAchievement: Starstruck

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: YOLOAchievement: Starstruck

Block or report dialectforge

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
dialectforge/README.md

Hi, I'm John Martin (DialectForge)

I'm a carpenter. I build software by directing AI coding assistants (Claude and Grok). DialectForge LLC is my one-person shop.

Projects

  • FlowState: many AIs, one memory. A free, local-first MCP server that gives Claude, ChatGPT, Cursor, Grok and any MCP app one shared project memory in SQLite on your own machine. You decide what gets saved. MIT. Latest release (v2.0.1) | Glama listing | More

    FlowState on Glama

  • Lu: an experimental, persistent AI written in Rust. It's built on a knowledge graph instead of model weights, and it's designed to say "I don't know" and to remember its mistakes ("scars"). Research code. AGPL-3.0, with a commercial license available. More

  • Cattledog: a small spinoff of Lu that wraps an LLM with Lu's honesty and scars. It blocks tool calls that match remembered forbidden actions. Python + Rust. Apache-2.0. More

  • DF-VPN (in progress): a censorship-resistant VPN built on the DialectForge protocol. The service is offline while the code is prepared for open-source release. More

Testing note: All testing so far was done by me with AI help; nothing has been verified by anyone outside yet.

An idea, not built yet

How Lu, Cattledog, Jev and LLM agents might fit together (write-up). Jev is a third-party model from TypeSafe AI, not made by me.

flowchart LR
    U[User task] --> A[LLM agents]
    A <--> J[Jev: fast typed decisions]
    A <--> C[Cattledog: honesty + scars]
    A -. thinking, inputs, outputs .-> L[Lu: persistent, curious]
    J -. decisions .-> L
    C -. scars .-> L
    L -- guides --> A
Loading

For your AI

Your AI can read everything: AGENTS.md in each repo, llms.txt on the site. Vibe coders are welcome.

Links

Pinned Loading

  1. cattledog cattledog Public

    Cattledog: experimental honesty and "scars" memory layer for LLM agents. Says "I don't know" when an answer is not grounded and blocks tool calls that match past mistakes or forbidden actions. Pyth…

    Rust

  2. lu lu Public

    Lu (Lu-4.0): experimental, persistent learning AI built on a knowledge graph instead of neural network weights, written in Rust. Designed to say "I don't know" and to check actions against memories…

    Rust

  3. FlowState FlowState Public

    Free, open-source, local-first project memory for your AI apps (MCP server for Claude, ChatGPT, Cursor, Grok and any MCP client). Your data stays on your machine.

    Python 22 2