Runtime Surface Map
AxonFlow has several integration surfaces because production AI systems are not all shaped the same way. The confusion usually starts when a team tries to compare framework integrations, Workflow Control Plane, Multi-Agent Planning, MCP connector governance, and Decision Mode as if they were alternatives on one axis.
They are not. They answer different questions.
The Short Version
| Surface | Primary question it answers | Use when |
|---|---|---|
| SDK gateway mode | "Can my existing app ask AxonFlow before and after an LLM call?" | You want to keep provider calls in your app or framework. |
| SDK proxy mode | "Can AxonFlow own the governed request lifecycle?" | You want one runtime endpoint for provider routing, policy, redaction, cost, and audit. |
| Framework integrations | "How do I wire AxonFlow into LangChain, LangGraph, CrewAI, Semantic Kernel, or similar tools?" | Your app already uses an agent or orchestration framework. |
| Workflow Control Plane (WCP) | "Can an external workflow engine ask for step gates, approvals, retries, and audit?" | Your workflow runtime stays outside AxonFlow but needs governed step boundaries. |
| Multi-Agent Planning (MAP) | "Can AxonFlow generate, store, execute, and inspect multi-step plans?" | You want AxonFlow to manage the planning lifecycle, not only gate an external workflow. |
| MCP connector governance | "Can AxonFlow govern tool calls and connector outputs?" | Agents or plugins call databases, SaaS APIs, storage, or HTTP tools. |
| Decision Mode | "Can my infrastructure gateway use AxonFlow as a PDP?" | You already operate agent, MCP, or LLM gateways and want centralized decisions. |
How To Think About The Axes
There are three separate decisions:
- Who calls the model or tool? Your app, AxonFlow, an external orchestrator, an MCP connector, or an infrastructure gateway.
- Who owns execution state? Your app, the external framework, AxonFlow MAP, or WCP workflow records.
- Where should governance happen? Before a prompt, after a response, before a tool call, at every workflow step, or at a gateway boundary.
Most real deployments combine more than one surface. A LangGraph app might use the LangGraph integration for developer ergonomics, WCP for step gates, MCP governance for database tools, and Decision Mode later at an enterprise gateway.
Surface Boundaries
SDK modes
Use SDK modes when the application is the integration point.
- Gateway Mode keeps your provider call path in your code.
- Proxy Mode sends the governed request through AxonFlow.
- Choosing A Mode compares the tradeoffs.
Framework integrations
Use framework integrations when your app already has a runtime shape from a framework.
Framework pages should show how AxonFlow fits into the framework's natural execution lifecycle. They should not imply that the framework itself becomes the governance system.
Start with:
Workflow Control Plane
WCP is a governance plane for external workflow runtimes. Your orchestrator still executes the workflow. AxonFlow records workflow state, evaluates step gates, captures audit evidence, and handles approval-aware transitions.
Use WCP when:
- workflow execution already lives in LangGraph, LangChain, CrewAI, n8n, or an internal engine
- step-level gates matter more than single-request policy checks
- retry, checkpoint, approval, and audit semantics need to be explicit
Start with Workflow Control Plane Overview.
Multi-Agent Planning
MAP is the AxonFlow-managed planning lifecycle. The platform can generate a plan, store it, execute it, expose execution state, and apply governance along the way.
Use MAP when:
- you want AxonFlow to manage the plan lifecycle
- agent definitions and planning patterns are part of the platform design
- you need execution visibility without owning a separate orchestration framework
Start with Multi-Agent Planning Overview and Getting Started with MAP.
MCP connector governance
MCP governance is about tools and data systems. It is orthogonal to whether the model call uses gateway mode, proxy mode, WCP, or MAP.
Use MCP governance when:
- agents call databases, storage buckets, SaaS systems, or HTTP APIs
- connector input and output need policy checks
- response redaction and exfiltration controls matter
Start with MCP Overview and MCP Policy Enforcement.
Decision Mode
Decision Mode is for teams that already operate gateway infrastructure and want AxonFlow to act as a policy decision point.
Use Decision Mode when:
- many applications sit behind common gateway layers
- changing every application is slower than enforcing policy at the gateway
- audit and trace correlation across gateway layers matter
Start with Decision Mode.
Common Combinations
| Architecture | Common combination |
|---|---|
| New AI app with one request/response flow | Proxy mode + provider routing + audit |
| Existing LangChain or LangGraph app | Framework integration + gateway mode or WCP |
| Agent with database tools | Framework integration + MCP connector governance |
| Long-running workflow with approvals | WCP + HITL approval gates + checkpointing |
| AxonFlow-managed multi-step plan | MAP + workflow APIs + execution viewer |
| Enterprise gateway estate | Decision Mode + gateway adapters + audit correlation |
| Regulated production rollout | Runtime surface + Trust Center + compliance mapping + evaluation license |
Common Misconceptions
This page exists to prevent three mistakes:
- treating WCP and MAP as duplicates
- treating framework integrations as security boundaries by themselves
- treating MCP connector governance as only an integration feature instead of a tool-risk control
If you are still unsure, start with Runtime Request Paths. It tells you which endpoint family to call first. This page explains how the product surfaces fit together.
