Tourbillon.ai — Precision AI Solutions

01 Products

A protocol that unlocks
the power of your agents.

One open language at the center, a growing family of instruments around it — every token, every call, every agent earning its keep.

Flagship Released · v0.2.8 · Apache-2.0

TERSE

Token-Efficient Representation Semantically Expressed

An open hierarchical state language for humans and AI agents — readable like notes, queryable and patchable like data. Where JSON taxes every key and free prose resists safe mutation, TERSE sits between: the file is the state, and agents patch it by declaration instead of rewriting whole documents. The result: ~2.5× more token-efficient AI calls and responsibly less datacenter load per query.

See it in action → Read the spec GitHub

Officially released — full spec, written guide, and syntax reference
Query & mutate by name: ? Cases.* [WHERE Its.priority == "high"]
Silence preserves — what you don't restate, you don't change
Apache-2.0: any implementation, commercial or otherwise, can build on it
Reference implementation in Python (terse-py); TypeScript next

state.terse — the file is the state● LIVE

JSON · 122 tok

{
  "cases": [{
    "id": 4412,
    "title": "outage",
    "priority": "high",
    "status": "open",
    "tags": ["billing", "auth"],
    "reporter": {
      "name": "Ada Reyes",
      "phone": "555-0100"
    },
    "assignee": {
      "name": "Miguel Ortiz",
      "team": "platform"
    },
    "summary": "Night shift lost auth
      to the billing API. Escalated
      to on-call at 02:31."
  }]
}

TERSE · 49 tok

# Cases
## 4412 outage(priority: high; open; billing; auth)
reporter(name: "Ada Reyes"; phone: "555-0100")
assignee(name: "Miguel Ortiz"; team: platform)
"""
Night shift lost auth to the billing API.
Escalated to on-call at 02:31.
"""

// same state — now query & patch it by name:
? Cases.* [WHERE Its.priority == "high"]
## 4412 outage(+resolved; -open)

JSON

122 tok

TERSE

49 tok

Same meaning. Same machine-readability. ~2.5× fewer tokens

§1

Tokens are a resource

JSON taxes every key; free prose resists safe mutation. TERSE sits deliberately between — sparse enough to keep in context, structured enough to trust.

§2

The file is the state

Readable like notes, queryable and patchable like data. No translation layer between what humans read and what agents act on.

§3

Silence preserves

Agents patch by declaration — what you don't restate, you don't change. Mutations are surgical, idempotent, and audit-friendly.

// The TERSE family

Enterprise Coming soon

TERSE/DB

TERSE state storage & protocol at enterprise scale

TERSE/DB is the TERSE state storage and protocol system implemented at enterprise-class scale, reliability, and performance — and it's 100% compatible with local TERSE applications for agents. When AI state grows beyond single files and needs to be shared across multiple agents and AIs across the enterprise, TERSE/DB is the solution.

Contact sales → +1 805-416-0415

Deployable in a network or cloud environment — managed or customer
Built-in redundancy modes and data recoverability
Terabyte-scale state namespaces with retrieval size guards
Transaction-level support, up to full serializability
Native encryption-at-rest and transport protocols
Full security and access control model

Released · Apache-2.0

terse-brain

a self-updating librarian for your knowledge

A personal knowledge base an LLM incrementally builds and maintains — an AI that continually reads, organizes, and compounds your documents, acting as a self-updating librarian. It's the 1:1 port of Karpathy's llm-wiki pattern to TERSE: the same brain, with every bookkeeping structure typed, queryable, and linted for free.

terse-brain on GitHub ↗

terse-brain diagram — 301 nodes · 432 edges · one brain.terse file

TERSEbrain interface: an interactive knowledge graph of raw sources, entities, concepts, claims, and syntheses arranged in concentric rings, with search, filters, and a node inspector

The index is computed — ? [DEPTH 2; CONTAINERS] — never hand-written, never stale
Every claim must cite a real source; provenance is hashed at register time
References are parser-known — dangling links surface as lint findings, not rot
Karpathy's failure mode, drift, becomes six lint checks you can run any time
Ingest is one transactional call: wiki upserts, contradiction flags, log entry
100% local, 100% yours — one file, one CLI, no service dependency

// benchmark — terse-brain vs Karpathy llm-wiki, out of the box · same 15 docs, 72 questions, 3 reps · same model & independent judge

1:1

Same brain — accuracy at parity

6.5^×

Fewer tool calls — 109 vs 709

3.5^×

Faster — 23 min vs 80 min

2.9^×

Cheaper — $2.10 vs $6.09 / run

Released · Apache-2.0

terse-memory

Durable agent memory as a TERSE store — no embeddings, no pipelines, no vendor black box. One file you can open, diff, and edit with the same CLI your agent uses; every MCP-speaking host reads and writes the same bytes. The product is a prompt; the store is the framework.

terse-memory on GitHub ↗

Plugin

CodeBoss

Orchestrates Claude Code headless from Cowork. Cowork acts as the supervisor with full UI access; Claude Code runs silently in the background and messages back when the work is done. Delegate the build, keep the oversight.

github.com/lrh-tourbillon/code-boss ↗

02 Get started

Three ways in.

TERSE is a language and a protocol, not a framework — start at whichever layer you're standing on.

/01

In a prompt

Zero install. Paste the written guide into any system prompt and keep state in TERSE — works in a chat window, works in every framework.

paste terse-spec/TERSE.md
→ your system prompt. done.

Start here →

/02

In your code

The Python reference implementation, the shared MCP server, and the products on top — one clone, all Apache-2.0.

$ git clone github.com/terse-lang/terse
$ pip install -e ./terse-py -e ./terse-mcp

Install guide →

/03

In your agent stack

One MCP server, every host — plus integration patterns for LangGraph, OpenAI Agents SDK, Microsoft Agent Framework, Google ADK, LlamaIndex, and CrewAI.

$ terse-mcp map brain ./brain.terse
# every MCP host reads the same bytes

Integration patterns →

// The full guide, with real install commands and the framework grid — Getting Started.

03 Services

AI transformation,
end to end.

From first roadmap to production systems — for small companies through large enterprises.

/01

Strategy & Roadmapping

Assess where AI creates real leverage in your business — and sequence the work so value lands early.

/02

Integration Engineering

Wire models into the systems you already run: ERPs, CRMs, data warehouses, internal tools.

/03

Custom AI Systems

Purpose-built agents, pipelines, and applications designed around your workflows, not ours.

/04

Data Readiness

Clean, structure, and govern the data your models depend on before it becomes the bottleneck.

/05

Governance & Compliance

Policy, audit trails, and controls so your AI footprint stands up to scrutiny — internal and regulatory.

/06

Enablement & Training

Your team, fluent in the tools. We build capability, not dependency.

// Full service catalog in progress — contact us to scope an engagement.

TOURBILLON // MECHANISM EST. PRECISION

04 About

Named for the precision mechanism that refuses to drift.

The first wave of AI was about proving it could be done. The new wave is about doing it well: systems that are efficient by design, accountable by default, and measured in outcomes rather than demos. We build products that strip waste out of every model call, and we deliver the engineering that gets organizations — from small companies to large enterprises — from AI ambition to AI in production.

Less spectacle. More mechanism.

We built Tourbillon because we were inspired by its real-world counterpart: a watchmaking "complication" invented to cancel out the errors gravity introduces — a mechanism whose entire purpose is precision under real-world conditions. That's the standard we hold our work to.

// Founders

Lou Houlemarde

Co-Founder & CEO

A former EVP at Trusaic, Lou is a prolific inventor, entrepreneur, and AI thought leader with deep expertise in AI systems, cognition, and transferability. He is the inventor of the TERSE syntax — the language at the center of everything Tourbillon builds.

Daniel Houlemarde

Co-Founder & Chief Product Officer

A former VP of Data & Products at NBC News, Dan has built business and consumer experiences serving 100M+ users — harnessing the power of emerging AI to turn data into products people rely on every day.

Built for the era
of efficient AI.