Our AI watch infrastructure
Every day, it tracks how AI is evolving, qualifies useful signals and feeds the work we do for clients: our judgement, our AI projects and our long-term guidance.
No longer commercialized · Still in active use5 to 10
qualified signals / day
AI signals that matter before they become obvious.
In plain terms
Watch
Spot
5 to 10 signals / dayQualify
Summaries · key ideas · knowledge baseA continuously fed reading
Built to last
18
months of development
4,144
automated tests
4
dedicated agents
11.5B
tokens used
374,228
lines of code
These numbers say one simple thing: the InsightKeeper agent is not a demo. It is a maintained, used, stress-tested product. Specialized coding agents and agentic engineering practices are what make it possible for two founders to keep a product of this size alive without losing quality or pace. The agents make the work sustainable, they do not decide it for us.
Four operational roles, each framed to keep the product running in production.
Monitors logs, fixes trivial issues and prepares an actionable diagnosis when human arbitration is required.
Updates dependencies fast enough to absorb useful improvements, never so fast that it weakens what already works.
Keeps internal documentation usable by other agents: precise enough for onboarding, compact enough to remain token-efficient.
Improves our in-house models and heuristics from observed dysfunctional uses and recurring product patterns.
At the origin
The InsightKeeper agent first existed as a consumer Mac and iPhone app, designed to turn the daily flow of digital content into a usable knowledge base. Articles, podcasts, videos, posts, preprints: everything could be captured, qualified by local AI, summarized, indexed and found again without friction. The goal was simple: move faster from raw information to useful knowledge.
We chose to stop commercializing it for the general public. The product now serves our work better in another form: as internal infrastructure for watch, R&D and AI practice. That product history matters because it forced us to handle real AI problems until the application was ready to be sold on Apple’s store.
What this changes for you
The agent does not replace our judgement. It feeds it. It helps us better distinguish what is becoming possible, what remains fragile, what deserves a test and what is mostly noise.
We position your ideas against what we actually observe: emerging uses, technical limits, field feedback, hidden costs and tool maturity.
Our training evolves from real cases and concrete experiments. Thanks to our watch agent, we identify more emerging topics, and AI lets us test their value faster.
Our watch and R&D help us recognize earlier which topics are becoming accessible for your business tools, modernizations and AI systems.
The InsightKeeper app is no longer offered to new users and has been removed from our offer. The InsightKeeper agent remains in active use as a strategic internal asset for our firm.
Working with companies in France
We work closely with leadership teams to clarify AI priorities, structure useful projects, and execute them seriously. That proximity is central to how we deliver value.
