Fashion
ÉLAN
- You give One photo.
- Expert Core ÉLAN
- Delivers A style diagnosis, item matching and a guided recommendation.
Different work. The right Expert Core for each task.
Start with the goal or source material. Each Core applies domain-specific judgment and returns a reviewable professional output.
Interior
The Designer
- You give One blank 2D floor plan.
- Expert Core The Designer
- Delivers A reviewable space plan, design direction and visual proposal.
Film
The Director
- You give The product, or just the goal.
- Expert Core The Director
- Delivers Creative direction, story structure, storyboard, shot design and a production-ready plan.
Your domain
One key. Every expert.
The three above are Cores we have built and delivered. They sit behind the same key, and the task decides which one answers — you do not pick.
Talk about your CoreAny agent
Your Agent.
Keep the model and runtime you already use. One key adds the layer above them — nothing gets replaced.
One key. Two stacks.
Persistent Memory keeps the user’s context consistent. The SHIKI Expert Core brings authorized human judgment to the task.
SHIKI
argsmem
SHIKI Expert Core
SHIKI
Authorized human-derived judgment for the task: what to notice, how to decide, and what good looks like.
You giveA domain expert, real completed work, and the quality standard that must hold.
You getA Core your agents can call — expertise, not memorised answers.
Persistent Memory
argsmem
Governed user context across agents, models and sessions — what happened, what changed, and what must remain true.
You giveThe systems it may access, the boundaries it must respect, and feedback from real work.
You getContext that survives every handoff, learns from reviewed outcomes, and stays auditable.
Together
SHIKI routes work to human-derived intelligence.
Model routers send prompts to models. This routes the work itself to authorized human judgment — while the user’s governed Memory stays consistent underneath.
Questions about the work
What does an Expert Core actually contain?
Domain knowledge, the decision criteria an expert applies, the working methods they follow, and the boundaries they will not cross. It is built from authorized source material — real decisions and completed work — not from a written brief about how the expert behaves.
How is an Expert Core different from a system prompt or a custom agent?
A prompt describes a role. A workflow fixes the steps. An Expert Core carries how a professional evaluates a problem, what they weigh against what, and where they stop — so the judgment holds on work the prompt never anticipated.
Who decides what the agent is allowed to see and keep?
You do. Memory operates inside the permissions and project boundaries you set. Approved decisions, corrections and evidence stay reviewable, and short-term signals cannot silently overwrite an established rule.
How does this connect to the agents and models we already run?
Through one key, above your existing stack. Your models, tools and runtime stay as they are — the key adds the expert judgment and the continuity the agent is missing.
What happens to what the agent learned when we change models?
It stays. Memory and the Expert Core sit above the model layer, so replacing or upgrading a model does not reset the context or the standard the work is held to.
How do we know the upgrade worked?
Each engagement starts with one workflow, a defined expert standard and agreed success measures. The result is validated against those measures through reviewed output, not asserted.
Start with one workflow
Upgrade the agent you already use.
Bring us one workflow that cannot afford generic judgment. We will define the first measurable proof with you.