Composable
Skills, context and permissions assembled per role, not per agent. Swap a skill, version a context graph, tighten a permission boundary: the Hat recomposes and every running agent picks up the change without a re-deploy.
For CIOs & CISOs
Context-as-a-Service for your AI agents.
skilder decouples know-how from the AI agent. Your knowledge becomes a service; the model, an implementation detail.
- Tools : MCP endpoints from your stack (ERP, CRM, docs, internal APIs) plus generic utilities. Called on demand, no data replication.
- Skills : Instructions, executable scripts and reference appendix in one package. Versioned, signed, registry-stored.
- Hat : Composable unit: skills + tools + permissions for a role. The deployment primitive.
- AI agent : Any MCP-compatible client (ChatGPT, Claude, Copilot, in-house). LLM routed per task: cost, latency, sensitivity.
In their words
Trusted by builders.
A single agent can switch Hats depending on context: customer support on one request, contract analyst on another, without touching the code. I stopped thinking in dedicated agents; I think in reusable capabilities.
AI Architect · Legal · United States
Before skilder, deploying a new agent meant reconfiguring everything from scratch; tools, instructions, business context. With Hats, I define once, and any agent is operational in a single step. I went from 20 operations per agent down to one.
AI Engineer · Fintech · Switzerland
The technical primitive
The Hat Primitive. Composable, governed, portable.
One unit that bundles skills, context and permissions for a role. Deployable on any AI agent.
Governed
Every Hat has an owner, a workspace, an audit trail and a permission contract. Your CIO sees who deployed what to whom; your CISO sees what each Hat can read, write or call. AI Act and GDPR ready by design.
Portable
The same Hat ships to Claude, OpenAI, Microsoft Copilot or any MCP-compatible runtime. Build the role once, deploy across your agent stack. No rewrite per framework, no runtime lock-in.
Three steps, no code
How it works.
- 01
Co-create your roles
Your subject-matter experts build Hats in the Studio, with or without our authors. No code, anchored in your existing practices.
- 02
Connect your business systems
ERP, CRM, knowledge base, files, internal APIs, MCP. Your Hats access the right data, at the right moment.
- 03
Ship to your teams
SSO, roles, quotas, audit. Your people wear their Hats from Claude, ChatGPT or Copilot, the tool they already use.
Key features
01
Hat Editor
Composable agent persona: skills, context, and permissions bundled into a single deployable unit. One Hat, many agent frameworks.
02
Skills & Tools Catalog
The library Hats compose from. Skills (instructions + scripts + appendix) and tools (MCP endpoints): versioned, discoverable, governed. Portable to any compatible agent.
03
Context Graph
Your enterprise knowledge as a typed graph: vocabulary, processes, documentation, internal tools. Queryable by every Hat.
04
Framework Distribution
Deliver Hats to the agent framework your teams already use (Claude, OpenAI, Copilot, OpenClaw), without rebuilding for each runtime.
05
Workspace Governance
Multi-tenant isolation, role-based access, audit trail. Hat deployment is the permission boundary.
Tech stack
- LLMs
- Claude · GPT · Gemini · Azure OpenAI · Ollama (via Vercel AI SDK)
- Knowledge graph
- Dgraph · GraphQL (Apollo Server + Fastify)
- Skills runtime
- MCP server + client · AAIF Agent Skill standard · NATS JetStream event bus
- Identity & access
- OAuth (Google, GitHub) · workspace RBAC · NATS nkeys + JWT for runtime authentication
- Tenancy & data
- Workspace-scoped isolation · soft-delete lifecycle · idempotent schema migrations
Security & compliance
skilder is built for European enterprise constraints. Not around them.
| Deployment | Sovereign cloud hosted at Infomaniak (Switzerland). On-premise available on request via Docker images. |
|---|---|
| Compliance | GDPR by design. Full logging, exports, and right to erasure built into the runtime. |
| AI models | LLM-agnostic. Claude, GPT, Gemini, Azure OpenAI, Ollama: routed by task and sensitivity. |
| Encryption | TLS in transit, encryption at rest, no training on your data. |
From scattered prompts to governed, shared roles.
Try skilder (no credit card), or scope your use case with our founders.
Supported by
