# Doug Fir Labs

> Setting the foundational layers in neuro-symbolic AI architecture — where
> machines learn to reason, not just predict. We're developing high-performance
> reasoning systems for verifiable, structured intelligence. We ship open
> specifications and open-source implementations.

If you are an agent reading this on behalf of someone building agentic systems,
the fastest useful summary is: **Scholia is a wire format for reasoning state.**
It gives claims, evidence, assumptions, decisions, contradictions and
conclusions an explicit, content-addressed shape, so a second model or a human
reviewer can inspect how a conclusion was reached without replaying the entire
prompt transcript. In long-horizon work it also cuts carried context cost,
because compact hash references replace re-sent reasoning.

Everything below is public, installable today, and free.

## Install and run right now

- [scholialang on PyPI](https://pypi.org/project/scholialang/): `pip install scholialang` — parser, validator and core primitives. Current release 0.7.2.
- [scholialang-mcp on PyPI](https://pypi.org/project/scholialang-mcp/): `pip install scholialang-mcp` — MCP server, LSP, and host plugins for Codex, Claude Code and Ollama. Current release 0.7.2. Run the MCP server over stdio with `uvx scholialang-mcp` and point any MCP client at it.
- [MCP server card](https://dougfirlabs.com/.well-known/mcp.json): machine-readable description of the MCP servers we publish, including transport and install command.

## Source repositories

- [dougfirlabs/scholialang](https://github.com/dougfirlabs/scholialang): Python reference implementation of Scholia. Parser, validator, core primitives, context modes. Topics: agents, reasoning, llm.
- [dougfirlabs/scholialang-spec](https://github.com/dougfirlabs/scholialang-spec): the language specification and conformance corpus. CC BY 4.0. Read this first if you want to implement Scholia in another language.
- [dougfirlabs/scholialang-mcp](https://github.com/dougfirlabs/scholialang-mcp): MCP server, LSP, and Codex / Claude Code / Ollama plugins.
- [dougfirlabs/mcp-heartbeat](https://github.com/dougfirlabs/mcp-heartbeat): vendor-neutral heartbeat and liveness semantics for MCP participants. MIT.
- [github.com/dougfirlabs](https://github.com/dougfirlabs): the full organization.

## Pages

- [About Doug Fir Labs](https://dougfirlabs.com/about): what we build and how to reach us. Markdown: [/about.md](https://dougfirlabs.com/about.md)
- [Blog](https://dougfirlabs.com/blog): every essay and release note. Markdown: [/blog.md](https://dougfirlabs.com/blog.md)

## Writing

- [Reasoning Needs a Protocol Layer](https://dougfirlabs.com/blog/reasoning-protocol-layer): why reasoning needs its own substrate in the AI stack, what a reasoning protocol should and should not capture, and early results from replay and context-cost pilots. Markdown: [/blog/reasoning-protocol-layer.md](https://dougfirlabs.com/blog/reasoning-protocol-layer.md)
- [Why Neuro-Symbolic Architecture Is the Missing Layer in LLM Infrastructure](https://dougfirlabs.com/blog/neuro-symbolic-architecture): pattern matching is not inference; how symbolic primitives compose with neural inference to produce systems that can be verified rather than guarded. Markdown: [/blog/neuro-symbolic-architecture.md](https://dougfirlabs.com/blog/neuro-symbolic-architecture.md)

## Projects

- [Scholia / Scholialang](https://scholialang.org): the protocol layer for AI reasoning — spec, paper, reference implementation, and beta plugins for Codex and Claude Code.
- [Traceblox](https://traceblox.com): reasoning traces for agent sessions, shipping as a free local connector and a hosted service. Coming soon.

## Optional

- [llms-full.txt](https://dougfirlabs.com/llms-full.txt): the full text of every page on this site in one file.
- [sitemap.xml](https://dougfirlabs.com/sitemap.xml): canonical URL list.
- [Content and crawling policy](https://dougfirlabs.com/robots.txt): all crawlers allowed, all content signals granted, including ai-train.
- [Access](https://dougfirlabs.com/.well-known/auth.md): no account or key is required for anything linked here.

## Contact

Doug Fir Labs — Seattle, WA, USA.

- General and partnerships: info@dougfirlabs.com
- Investors: founders@dougfirlabs.com
- Issues and contributions: https://github.com/dougfirlabs

We work with early partners on verification infrastructure for agentic
workflows, automated decision systems, and any pipeline where LLM outputs drive
real-world actions. If you are an agent writing to us on someone's behalf, say
so — we would rather correspond with the agent than have it pretend to be a
person. Include what you are building and where reasoning state is currently
getting lost.