Every agent action,
governed.
Every decision,
on the record.
Zograph is an agent platform with governance in the execution path. Build and operate specialist agents while keeping your options for models and deployment, with a clear record of consequential actions and the rules behind them.
Not ready to talk? Join the waitlist.
Builder Layer
Conversational and background agents from configuration: a prompt, a model allowlist, and plugins. Versioned draft to publish.
Read more →Compose
Bring your knowledge, tools and skills together.
Govern
Control, observe and prove every action.
Run
Keep your options for models and deployment.
Built for trust
Agent capability with organizational control.
Build
Layer 01 / 08
Compose the agent your work needs.
Build conversational and background agents in Zograph Studio.
Define their instructions, attach knowledge, MCP servers, plugins and skills, and choose the capabilities and model they need.
See Zograph in your world.
Illustrative examples
Choose an example to update the agents, tools and workflows throughout this page.
Attach your resources
Knowledge
Give the agent access to your content.
MCP Servers
Connect to your systems through MCP.
- Inventory SystemStock levels, orders, locationsconnected
- ERPProducts, suppliers, purchase ordersconnected
- Service DeskIncidents, changes, approvalsconnected
Plugins & Skills
Add reusable skills and tool packages.
- Data AnalysisTables, charts, insightsconnected
- Document GenerationWord, PDF, structured outputconnected
- Custom SkillsYour organization’s packages
Your choice of model
Choose from supported providers and compatible endpoints.
- Anthropic
- OpenAI
- Mistral
and more
Agent
Inventory planning agent
Helps planners resolve stock exceptions.
Instructions
You work with our inventory data to identify and investigate stock exceptions, propose actions, and prepare a briefing for the planning team.
- Knowledge
- MCP Servers
- Plugins & Skills
- Add
Capabilities shown in this example
Task Planning
Let the agent plan and manage complex work.
- Analyze stock exceptions
- Investigate root causes
- Propose resolution options
- Prepare briefing
Artifacts
Let the agent create and iteratively refine documents.
Stock exception memo
DOCA structured briefing including key findings, impacted SKUs, root causes and recommended actions.
Two ways to put agents to work.
Conversational agents
Work with an agent.
Ask questions, review its work and guide the next step through conversation.
Inventory Planning Assistant
You“Help me review this stock exception.”
Agent“Here is the inventory context and a draft exception memo.”
Background agents
Delegate work to an agent.
Run agents on a schedule or trigger them from your systems. No open conversation required.
Started by
More ways to equip your agent.
Memory
Keep useful context across conversations.
Learned Heuristics
Apply reviewed lessons from earlier work.
Web Tools
Search the web and retrieve pages.
Code Sandbox
Run code in a sandboxed environment.
Feedback
Collect feedback on agent replies.
Suggested follow-ups
Suggest the next message in a conversation.
More capabilities are in development.
Builder / 01
02 / Evaluation
Test behavior.
Before release.
Run the scenarios that matter to your business.
Inspect the response and evaluation results.
Example evaluation
Test scenario
Stock exception briefing
Investigate the exception and prepare a briefing using the connected inventory data.
Agent response
Stock exception memo
The stock position, supporting evidence and questions for the planning team.
Evaluation checks
- passRequired sections
- passSources included
- failUncertainty flagged
Refine the agent. Run the scenarios again.
03 / Extensibility
Your logic.
In the lifecycle.
Extend Zograph with synchronous HTTP hooks.
Connect your systems through event subscriptions.
Synchronous hooks
Call your service and use its response.
Event subscriptions
Let your systems react to platform events.
Your services, connected to the platform lifecycle.
Evaluate + Extend / 02
Govern
Layer 04 / 08
See the work behind the result.
Inspect conversations and background runs, from individual tool calls to their sources and artifacts.
Understand execution, model usage and time spent waiting.
Inventory Planning Assistant
Stock exception SR-3184 · Example conversation
Recorded actions
Execution
Approval wait
Continue
- Read the exceptionget_exception
- Read stock levelsget_stock_by_dc
- Read the forecastget_forecast
- Read open purchase ordersget_open_purchase_orders
- Create purchase ordercreate_purchase_orderWaiting for approval
- Edit exception memoartifacts_str_replace
Illustrative execution · not a time scale
Associated context
Task planInventory proceduresException memoSelected action
Create purchase order
Recorded result
Purchase order created.
Inventory System · MCP tool
Permission and execution are separate facts.
Execution
Outcomes, duration and approval waits.
Usage
Models, tokens and estimated cost.
Context
Plans, sources and created artifacts.
Observability / 04
Set the rules. Control the action.
Set defaults across your agent’s tools, refine permissions by source or action,
and choose when approval is required.
Configure the boundaries
Broad defaults. Inherited settings. Deliberate exceptions.
Action
Effective permission
Deciding source
- Read inventoryInventory SystemAllowedAdministrator rule
- Update task planTask PlanningAllowedPlugin default
- Edit exception memoArtifactsAllowedUser standing grant
- Create purchase orderInventory SystemAsk administratorExplicit tool rule
- Change safety stockInventory SystemDeniedAdministrator rule
Governed action
Create purchase order
Example request
Effective permission
- Ask an administrator each time.
Initiated by
- Requesting user
The requested call is held until an administrator decides.
Rules establish permission.
Across plugins, provider tools and MCP servers.
Grants carry user authority.
Automatic allows still have a deciding source.
Your logic can participate.
Connect decision logic through an extension.
Access & Governance / 05
Every decision, with its reasons.
Keep the history of changes, tool permissions, attempted actions and approvals.
Trace an action back to the rule, authority and turn behind it.
Connected records
Changes
HistoryWho changed or published what.
Tool permissions
AdmissionsWhat was allowed, denied or required approval.
Invocations
What was attempted and how it ended.
Approvals
Who decided, and how the request was resolved.
Selected record · Invocation
Related turnCreate purchase order
Allowed following administrator approval.
Authority
Administrator approval required each time.
Explicit rule for this tool.
Approval
Approved by an administrator.
Decision recorded before tool execution.
Execution
Tool outcome recorded separately from permission.
Related context
Inventory Planning Assistant
Initiated by the requesting user.
Inspect the associated turnAutomatic decisions, too
Edit exception memo
Not only the exceptions. The automatic allows, too.
Records that support review and accountability.
Audit & Compliance / 06
Run
Layer 07 / 08
Choose your models. Keep your options.
Use Zograph’s model access, or bring your own provider credentials.
Choose from supported models and compatible endpoints, with control over the connections behind your agents.
Zograph-managed access
Model access supplied by Zograph.
No provider API key required from you.
Your own provider credentials
Connect through your own provider accounts, cloud platforms
or compatible endpoints.
One platform
Your agents, in control.
Inventory Planning Assistant
- Agent configuration
- Governance
- Execution records
Your platform stays the starting point.
Evaluate behavior when changing models.
Model connection options
Direct APIs
Connect directly using your provider credentials.
- AnthropicAvailable
- OpenAIAvailable
Cloud platforms & routing
Connect through the platform or routing service you use.
- Amazon BedrockAvailable
- Microsoft Foundry (Azure)Available
- Google Cloud / Vertex AIPlanned
- OpenRouterPlanned
Your inference endpoint
Connect a supported private or self-hosted inference service.
- Self-hosted modelsAvailable
- OpenAI-compatible endpointsAvailable
Additional integrations are planned. Tell us which platform your team uses.
Connection choices, not a failover sequence.
Model choice
Choose from the models your connections support.
Zograph access or your own keys
Start with Zograph-managed access, or connect your own provider account.
Compatible endpoints
Connect beyond a fixed catalog through supported interfaces.
Redundancy / 07
Your platform. Your environment.
Choose where Zograph runs and how it connects to your models and systems.
Keep deployment, regional choices and external dependencies explicit.
Platform deployment
Zograph-managed
Available regions
- European Union (EU)
- United States (US)
Platform
Your governed agent platform.
- Agents
- Governance
- Execution records
Connected services
Inference endpoint
Your configured model connection.
Placement depends on configuration.
Business systems
Inventory, knowledge and tools.
Example deployment. Model and business services sit outside the platform environment in this configuration.
Deployment choice
Where the platform is operated.
Regional choice
Where the deployment is located.
Explicit connections
Which models and systems the platform depends on.
Sovereignty / 08
Walk us through
your case.
Forty minutes with the people who built Zograph.
Start with the workflow you want to govern first.
Not ready to talk? Join the waitlist.

