Skills: The Capability System
Installable, versioned capability packages: marketplace, git import, and skills authored in chat. Each one is scanned before install and runs in a hardened sandbox with no network by default.
From platform to people.
The member-facing experience on top of the four modules: every signed-in employee works with their permitted Company Expertise, personal memory, documents, governed tools, skills, and scheduled tasks, in one workspace.
Every company now wants a ChatGPT that knows their business. The generic version of that promise fails three ways: it only knows what was written down, it answers without governance, and it cannot touch the systems where work actually happens.
Company GPT is that promise built on the four modules instead of a document dump. It answers from mined expertise and governed Company Expertise, respects what each member on each channel may see, acts through approved tools, and keeps improving as the Continuous Learning System returns approved context to it. It is never sold standalone: it becomes genuinely useful only after enough expertise has been mined.
Six properties define the member experience.
Signing in lands the member directly in the company's default member deployment: conversations, documents, tasks, and memory in one place, with no setup step.
Every answer draws on the member's permitted Company Expertise, personal memory, conversation history, and shared documents, under channel policy.
Capabilities install as governed skill packages: seeded document skills, a curated marketplace, git import, or skills authored in chat, every one scanned, sandboxed, and activated by a human.
Scheduled tasks, mid-conversation background delegation, and step-budgeted autonomous runs let members delegate work, not just ask questions.
Through 180+ integrations, MCP servers, and deterministic System State stores, the assistant creates tickets, updates records, and drafts messages, every write passing the approval protocol.
Roles and permissions are enforced server-side, and the whole experience deploys on-prem: self-hosted or air-gapped, with offline licensing, signed metering, and bring-your-own-key models.
One workspace, governed like every other surface.
Six systems make the workspace more than chat.
Installable, versioned capability packages: marketplace, git import, and skills authored in chat. Each one is scanned before install and runs in a hardened sandbox with no network by default.
Agent plus prompt plus schedule, run as the owning member with their permissions and connections. Background delegation posts a summary back; write actions suspend until a human decides.
Audience, roles, knowledge visibility, tool tiers, and connection scopes enforced server-side at every layer, not styled in the UI.
A 180+ integration catalog, per-tenant MCP servers for ERP and custom systems, data sync into workspace knowledge, and typed context schemas. Deterministic values are read exactly, never inferred.
Self-hosted and air-gapped installations are first-class: signed time-limited licenses, graceful degradation, offline hash-chained metering, and bring-your-own-key or self-hosted models.
Approval mode chip, inline approval cards for write actions, connect cards offered mid-conversation, and human handover identical to every other channel.
Behavior comes from the published agent, the channel's knowledge policy, and the member's identity and permissions, the same governance as every other surface. There is no side door.
It answers from mined expertise and governed Company Expertise, not just indexed files, so it knows the judgment behind the work, not only the policy PDF.
Skills, scheduled tasks, background runs, and write-tier tools make it a place to hand off work, with every action logged and approved.
Company GPT is offered once enough expertise has been mined, usually three to four meaningful expertise packages. If you want it first, the path starts with Expertise Mining.
Verti connects expert judgment, live context, governed action, and learning loops into one production system.