SpeakeasyvsMicrosoft
Agent 365 governs agents on Microsoft's path: Copilot Studio, Foundry, Work IQ MCP, and Intune-managed Windows. Speakeasy runs inside the agents your teams already use, on macOS, Windows, and Linux, as the same Entra principal. Then it governs the gateway too.
Which should you pick, Speakeasy or Microsoft Agent 365?
On an Entra tenant both govern the same Entra Agent ID principal, so the question is where it is checked. Pick Microsoft when your agents are built on Copilot Studio or Foundry and run inside Microsoft 365, when your fleet is Intune-managed Windows with Defender XDR as the security operations center, or when Entra Agent ID, Conditional Access, and lifecycle workflows are the whole requirement. Pick Speakeasy when AI is rolling out on macOS and Linux as well as Windows, when your teams run third-party agents, or when your tools sit behind your own APIs.
AI is rolling out across the whole company, on macOS and Linux as well as Windows, and you want the same Entra Agent ID principals governed in every agent and tool your teams already use.
Your agents are built on Copilot Studio or Foundry, run inside Microsoft 365, and your fleet is Intune-managed Windows with Defender XDR as the security operations center.
Agents run as Entra Agent ID principals synced into Speakeasy, or as Okta principals, with the grant checked on every call from any OS.
Agent identity is minted from a blueprint in Entra, with Conditional Access and lifecycle workflows. Enforcement runs where Microsoft is in the path.
Hooks inside the agent on the prompt, the tool call, the file read, and the shell command, plus six inspection layers at the gateway, on macOS, Windows, and Linux.
Defender evaluates tool invocations for agents on Microsoft's path, Purview brings inline DLP in preview, and Defender for Endpoint covers Windows. Coding agents on macOS and Linux sit outside it.
Your agents, your identity, your APIs. MCP servers are generated from the API contract your teams already maintain, with evidence to Drata, Vanta, and your SIEM.
Deepest for agents built on Copilot Studio and Foundry inside Microsoft 365 on Intune-managed Windows. Third-party agents join through the Agent 365 SDK and Registry Sync, in preview.
Microsoft capabilities are drawn from Microsoft Learn and Microsoft's product blogs as of September 2026, including features in public preview. We build Speakeasy, so read us as an interested party. Speakeasy syncs Agent Identity from Entra Agent ID; this page compares scope, not the directory. See how Agent Identity syncs from Entra Agent ID.
Who is the agent, at every step?
On an Entra tenant the two agree on who the agent is: Speakeasy syncs Agent Identity from Entra Agent ID. They differ on where that principal is checked: inside the agent on every OS, or where Microsoft is in the path.
Okta or Entra- Directory principal
- Grant checked per call
- Call stamped with directory state
Identity lives in
Your identity provider
- Directory identity attached inside the agent, on every step
- People and agents share one permission model
- A denied call opens the grant that would allow it
- Entra ID
- Agent identity minted from a blueprint
- Conditional Access for agents
- Tool invocation evaluated through Work IQ MCP
Identity lives in
Microsoft Entra
- Every registered agent gets an agent ID; Policy Templates and sponsor lifecycle workflows govern it
- Enforcement runs where Microsoft is in the path: Work IQ MCP, Copilot Studio, Foundry, Intune-managed Windows
You can only inspect what you can see.
Where Microsoft is in the path, Defender evaluates tool invocations before they run and Purview brings DLP. Speakeasy's hooks apply policy to the prompt before inference, the local tool call, the file read, and the shell command in Claude Code, Cursor, Codex, Cowork, and OpenCode on macOS, Windows, and Linux, and the gateway scans every MCP call.
- 01Hooks inside the agent: prompt, tool call, file read, shell command
- 02Server review with evidence dossier
- 03Prompt-injection heuristic
- 04LLM judge
- 05Natural-language guardrails, replayed before enforcement
- 06CEL detection rules
- 07Destructive-action detection
- 01Defender real-time protection on tool invocations, through Work IQ MCP
- 02Purview inline DLP (Copilot Studio, preview)
- 03Azure API Management llm-content-safety
- 04Defender for Endpoint runtime protection (Windows)
Prompts before the model sees them
Local tool calls, file reads, shell commands
Endpoint enforcement
Tool calls at the gateway
Agents outside Microsoft's path
Natural-language guardrails with replay
Server review at registration
Sensitive-data detection
XDR correlation and threat hunting
- Offered
- Partial, beta, or detect only
- Not offered or not documented
Govern what your team already runs.
Speakeasy governs the agents your teams already run, the directory you already have, and the APIs your teams already maintain, from the inside, on any OS. Microsoft's control plane is deepest for agents built on its platforms and devices managed by Intune.
- Your agents
- Your identity
- Your stack
- MCP servers generated from the API contract your teams maintain
- Catalog drawn from the official MCP Registry
- Coverage evidence to Drata and Vanta, events to your SIEM
- Agents
- Copilot StudioFoundryAgent 365 SDK
- Identity
- Entra Agent IDBlueprintsConditional Access
- Stack
- Work IQ MCPAzure API ManagementIntune (Windows)
- Deepest where agents are built on Microsoft platforms and run on Windows
- REST APIs become MCP tools through Azure API Management
- Third-party agents join through the Agent 365 SDK and Registry Sync (preview)
Speakeasy's AI control plane has been indispensable in enabling Fivetran's AI transformation.
Eli Davis
Fivetran
Questions
What is the difference between Speakeasy and Microsoft Agent 365?
Does Speakeasy work with Microsoft Entra Agent ID?
Does Microsoft govern Claude Code, Cursor, and Codex?
When is Microsoft a better choice than Speakeasy?
Why do hooks matter for governing AI agents?
More comparisons
Speakeasy vs Runlayer
A proxy sees the request. Hooks see the agent think: the prompt, the reasoning, the local tool call, the shell command.
Read the comparisonvsSpeakeasy vs Mintive
Both run hooks inside the agent. They differ on whose identity the loop runs as and who authors detection.
Read the comparisonvsSpeakeasy vs LiteLLM
A proxy sees the model call. Hooks see the agent think. The two run together, with LiteLLM kept as the model proxy.
Read the comparisonvsSpeakeasy vs Kong
An API gateway sees the request. Hooks see the agentic loop: prompts, file reads, shell commands, and local tool calls.
Read the comparison