The Causal Knowledge Company designs, builds and runs its own AI applications.
Regulatory assistants, financial diagnostics, media, tools for tradespeople: we turn LLMs, agents, knowledge graphs and causal reasoning into products people use every day — conceived, built and operated in-house.
One factory, several product families
Every app is stamped from a shared base — front-end, orchestration, retrieval, payments, voice — so each new product starts from what makes it unique: its corpus, its guardrails, its promise.
Our apps — built in-house, running in production
Regulatory assistants, financial diagnostics, media, tools for tradespeople, research: every application below is designed, built and operated by TCKC.
monassistant.pro
Three specialised French AI assistants: accounting, labour law, taxes — every answer cited.
A family of domain assistants that answer by voice or text and cite the official reference — chart-of-accounts number, statute article, tax-return box. Free questions at signup, then a 200-question pack with no subscription.
monassistant.pro
Three specialised French AI assistants: accounting, labour law, taxes — every answer cited.
monassistant.proFEC.NEWS
A French FEC file or tax-package PDF → full financial diagnosis, 100% deterministic numbers.
fec.newsBraquage Six
French GTA 6 news, verified in-game — news, guides and glitches with a clear status.
braquagesix.friArtiBot
You make the move, your bot does the rest — the tradesperson's assistant, paid per task.
iartibot.frHeliosForge
Stellar engineering argued through reproducible simulations — Dyson swarm, star lifting.
github.com/zepef/HeliosForgeParles au PCG
Voice-enabled French accounting assistant
parles-au-pcg.comParles aux Impôts
Voice tax assistant — 2026 French return
parles-aux-impots.comSix disciplines inside every app we ship
The same technical craft runs through every TCKC application — from the voice assistants to the financial engine.
LLM & multimodal intelligence
Model selection, guardrails and evaluation so every answer stays accurate, sourced and safe.
- Voice and text interfaces grounded in official corpora
- Expert-question benchmarks replayed at every change
- Guardrails: never an invented price, rate or legal reference
Agentic engineering
Agents doing the work behind our products, with deterministic fallbacks wherever reliability matters most.
- Automated monitoring and publishing pipelines
- User feedback triaged into tickets and draft fixes
- Human review where the stakes demand it
Knowledge graphs & Graph RAG
Retrieval that fuses structured knowledge and vector recall so answers stay truthful and explainable.
- Ingestion pipelines for official documents, APIs and SQL
- Every answer cites its reference: account, article, tax box
- Lineage from each answer back to its source
Causal reasoning
Causal models to simulate interventions, counterfactuals and scenarios before they happen.
- Cause-effect detection over heterogeneous data
- Branched scenario simulation
- Deterministic engines compute, the AI explains
Product craft
Design, full-stack engineering and operations working as one unit to ship usable software, fast.
- Design systems with accessibility baked in
- Web, mobile and fully on-device apps
- CI/CD with reproducible infrastructure
Trust by design
Safety, privacy and sovereignty built into each product rather than bolted on afterwards.
- Red teaming and jailbreak sweeps before release
- Data minimisation and French or on-device hosting where it matters
- Same inputs, same results: reproducible numbers
From an idea to a live app, in four moves
Every TCKC product follows the same rhythm — from the first signal to a live app we keep improving.
Signal
Each product starts from a real need in a domain we can master — a regulation, a niche, a trade. We secure the corpus and write down the promise the app must keep.
Prototype
A working prototype is assembled on the shared base, along with its guardrails and the expert questions it must answer before anyone else sees it.
Build & launch
The product ships on the factory's shared services — auth, payments, voice, emailing — with automated tests and quality passes at every change.
Run & improve
We run what we ship: telemetry, feedback triaged into fixes and, when the model works, a sibling product stamped from the same base.
An in-house platform that compounds product after product
Every tool built for one app is reused by the next — that is how a small team runs several product families.
Graph RAG Fabric
Retrieval pipelines that combine dense and sparse search, layer causal traces and surface citations automatically — the backbone of our assistants.
Agent Ops Console
Mission control for our agents: tool usage, cost, latency and guardrail scoring in one place.
Model Studio
Data curation, evaluation suites and distillation flows that turn a domain corpus into a specialised model.
Causal Playbooks
Scenario libraries and decision simulators to test an intervention before it reaches a live product.
All the right tools, orchestrated for momentum
The stack behind our apps — chosen product by product, shared across the factory.
Models & Intelligence
- OpenAI · Anthropic · Mistral
- Fine-tuning & distillation (QLoRA)
- Local inference (Ollama)
- Multi-provider orchestration with fallbacks
- Custom embeddings & retrieval
Knowledge & Retrieval
- Neo4j, Memgraph, RDF stores
- Weaviate, Pinecone, LanceDB
- Custom embedding distillation
- Causal graph compilers
- Temporal event stores
Engineering & Ops
- Next.js 16 — App Router, static & serverless
- LangGraph & Autogen
- Temporal, Airflow, Dagster
- Terraform + Pulumi IaC
- Great Expectations, Evidently
Proof, not promises
We prefer verifiable guarantees to testimonials: here is what our products promise — and hold — by construction.
“No figure ever comes out of a language model: a deterministic rules engine computes, the AI explains. Same files in, same results out, to the cent.”
Design principle — FEC.NEWS“Every answer cites its official reference: chart-of-accounts number, statute article, tax-return box. And every accounting entry is verified before display.”
Product guarantees — monassistant.proThe best proof is to try them: our live apps are one click away.
Try the appsA question about one of our apps?
Feedback, a bug, a press enquiry or a partnership idea: write to us and the team that builds the apps will read it.
