"Why not just use Anthropic?"
Use Anthropic. We do.
Anthropic supplies frontier models, Cowork, Claude Code, and agent runtime. GEIST uses Anthropic through its multi-provider gateway. Cowork covers the horizontal agent workspace. GEIST supplies tenant-aware domain execution: data, workflows, approvals, evidence, and deployment controls. Use both when a Claude session needs to perform accountable work inside a regulated business process.
Anthropic ships the horizontal AI layer.
GEIST puts it inside governed domain work.
Claude models
Reasoning and generation models available through APIs and first-party products. GEIST uses Claude for selected profiles behind provider routing and fallbacks.
Cowork
A cloud agent workspace with persistent projects, scheduled tasks, sub-agents, connectors, plugins, and cross-surface sessions.
Claude Code
Agentic software development across terminal, desktop, and IDE surfaces, with tool use and MCP integrations.
Managed Agents
Runtime primitives for custom agents, including sessions, environments, and server-side context. The builder still owns the application data model and domain workflow.
Enterprise controls
RBAC, spend controls, analytics, OpenTelemetry, and beta Compliance API coverage for supported Claude surfaces.
When a session becomes a regulated workflow.
Anthropic now supports organisations with access controls, connectors, scheduled tasks, analytics, OpenTelemetry, and compliance exports. Cowork is a governed horizontal agent workspace; GEIST does not duplicate it. GEIST owns the domain execution layer where a session must resolve a tenant, apply policy, update structured workflow state, coordinate human approval, and retain application evidence.
| Capability | Anthropic (Claude, Cowork, Code, or Managed Agents) | GEIST |
|---|---|---|
| Primary role | Frontier models, a horizontal agent workspace, and developer/runtime platforms | Multi-tenant domain applications and governed workflow execution |
| Persistent operational state | Persistent Cowork projects, sessions, files, instructions, and memory | Structured tenant-scoped records, work items, approvals, dependencies, and histories |
| Connectors and tools | Broad MCP connectors, plugins, and private enterprise marketplaces | Domain APIs and connectors executed under tenant policy with workflow state retained in GEIST |
| Organisation controls | Enterprise RBAC, groups, per-tool controls, spend limits, analytics, and OTel | Platform → Vertical → Instance → Tenant → User application scope and role enforcement |
| Long-running and scheduled work | Cloud Cowork sessions, sub-agents, and scheduled tasks | Application workers for syncs, ingestion, anomaly detection, review cycles, and domain workflows |
| Audit and evidence | Compliance API and OTel for supported Claude sessions and organisation activity | Domain actions, approvals, AI calls, cost, safety, and tenant attribution across GEIST workloads |
| Domain context | Customisable projects, skills, plugins, connectors, and agent instructions | Purpose-built schemas, regulated workflows, assessments, playbooks, and tenant-scoped knowledge |
| Provider strategy | Claude models and Anthropic-managed product surfaces | Tenant-aware routing across Anthropic, Vertex, deployment adapters, and local model paths |
| Deployment boundary | Anthropic-hosted product and model infrastructure | Customer Azure tenant, Hetzner, on-premise, or air-gapped deployment |
Product scope verified against Anthropic documentation and the GEIST runtime, August 2026.
A platform you couldn't recreate
with prompts.
AgentGEIST ships the AI Operating Model layer across 7 domains — strategy, use case discovery, governance, people & culture, engineering, data, and compliance — with pre-loaded assessments, ROI models, process maps, and scoped KBs in each.
13,720
ROI opportunities
8 industries × APQC × AI techniques
2,017
APQC processes
L2–L5 cross-industry taxonomy
4,769
BOD benchmarks
definitions + KPI measures
7
Operating-model domains
strategy, governance, people, engineering, data, compliance, use cases
Live voice agent
Talk to your platform.
Gemini 3.1 Flash Live with sub-100ms latency. Three conversation modes (interview, assessment, free conversation). Full transcript persistence and function-calling into the assessment + KB layer.
Reports, slides, diagrams
7 visual generation tools.
Visual reports, presentation decks, architecture diagrams, data-flow diagrams, swimlane process diagrams, assessment exports, and executive summaries — generated on demand from voice or chat. Every visual is grounded in your tenant's private knowledge — APQC process catalog, BOD benchmarks, organisational hierarchy, and 5-level scoped RAG — not a separate creative canvas.
7 functional areas
An entire AI Operating Model.
Strategy · Use Cases (Discover, Portfolio, Process Forensics) · AI Organisation · People & Culture (Capability Assessment, Resource Plan, Training, Change Mgmt) · Governance (Risk, Policies) · Engineering (Architecture, Partners) · Data (Connectors, Readiness, Quality, Workspace).
Assessment depth
7 pillars × 81 capabilities.
Our AI Maturity assessment (Gartner-derived) ships with 7 pillars and 81 tier-3 capability nodes. The AI Strategy assessment ships with 3 phases × 21 items. Both pre-loaded at install.
Every figure verified against shipped GEIST seed data — March 2026.
You're not choosing one AI product.
You're composing the right layers.
Use Cowork for horizontal knowledge work. Use Anthropic APIs or Managed Agents for custom agent experiences. Add GEIST when the workflow needs tenant-scoped domain state, policy, approvals, evidence, and deployment ownership.
Cowork + plugins
Horizontal agent layer
Rapid rollout for research, documents, recurring tasks, connectors, and team-specific agents.
Connect it to GEIST when work must enter a tenant-scoped domain workflow with durable evidence.
Custom application on Anthropic
Build the domain layer
Use Claude APIs or Managed Agents, then implement the data model, tenant policy, workflow state, integrations, evidence, and operations.
Maximum control; your team owns delivery, assurance, and maintenance.
GEIST + Anthropic
Compose both layers
Keep Claude or Cowork where they fit. Add pre-built tenant scope, domain workflows, approvals, evidence, deployment controls, and provider routing.
A governed domain application that complements Anthropic instead of attempting to replace it.
Commercial scope depends on seats, workflow depth, integrations, and deployment model. This is a product-boundary comparison, not a fabricated cost estimate.
Claude where it fits.
Provider control where it matters.
GEIST's gateway routes named profiles across Anthropic and Vertex with fallback chains and per-tenant attribution. The wider platform also supports Azure and Foundry adapters plus local model paths for deployment-specific requirements.
Anthropic is a material production provider, not merely development tooling. GEIST keeps model choice behind tenant-aware profiles so the application contract survives a provider or model change.
Multi-provider routing — current defaults
Anthropic
Primary for balanced and higher-reasoning profiles
Google Vertex AI
Primary for fast, economical, creative, and embedding profiles; fallback for Claude profiles
Azure / Foundry
Supported adapters for customer deployments and managed-agent integrations
Local models
On-premise paths for air-gapped and privacy-first deployments
Routing controls
Named profiles, fallback chains, eval-driven swaps, and per-tenant attribution
Verified against GEIST gateway defaults and provider adapters, August 2026.
Anthropic in production.
Optionality by design.
Production today
- ✓Claude is primary for balanced, best, premium, and reasoning profiles
- ✓Vertex fallbacks are configured for Claude-backed profiles
- ✓Prompt caching and extended thinking on supported models
- ✓5-level scoped RAG and reversible PII tokenization for governed AI workflows
- ✓Per-tenant attribution and metering across gateway-mediated workloads
Integration roadmap
- →Managed Agents runtime as a first-class provider — sessions, environments, agent definitions
- →Point GEIST at existing Managed Agent definitions and run them under tenant scope
- →Claude Files API as a knowledge source adapter
- →Batch API for offline processing pipelines
What deployment actually looks like.
Dermatology · Surgical AI
Wallderm
DermGEIST surgical decision support: wound assessment, ranked closure recommendations, voice-first OR workflow, and HIPAA-compliant image analysis via Vertex AI BAA.
Want to connect Anthropic to a governed GEIST workflow?
In a 30-minute architecture review, we map where Cowork or Claude should remain the user-facing agent and where GEIST should enforce tenant policy, workflow state, approvals, and evidence.