Speakeasy is now on the @JamfSoftware Marketplace. Deploy the Speakeasy Device Agent through Jamf Pro to govern every AI agent on your managed Macs.
marketplace.jamf.com/details/speakeโฆ
marketplace.jamf.com/details/speakeโฆ
0
0
9
Thank you to everyone who joined us and @Tailscale at The Allowlist to close out @aiDotEngineer NYC.
Put your MCP servers behind your Tailnet, then secure the rest of your enterprise AI usage with Speakeasy.
Put your MCP servers behind your Tailnet, then secure the rest of your enterprise AI usage with Speakeasy.
0
0
10
@aiDotEngineer NYC starts this afternoon.
Come play a round of shuffleboard at Booth 32 and tell us how your team is governing its agents.
Come play a round of shuffleboard at Booth 32 and tell us how your team is governing its agents.
0
0
9
Catch Sagar at @aiDotEngineer NYC next week.
"Permission to Act: Giving Agents Authority in Regulated Industries," Wed at 2:55pm ET.
Then come find us at the booth.
"Permission to Act: Giving Agents Authority in Regulated Industries," Wed at 2:55pm ET.
Then come find us at the booth.
0
0
12
We're hosting a happy hour with @Tailscale in New York on October 14. The Allowlist.
Most MCP traffic is internal, but the servers are exposed to from the public internet. Why? Speakeasy makes it trivially easy to move all your servers to your Tailnet. Put them on your private security of your AI workflows in seconds.
If you're in New York for @aiDotEngineer, come join. Drinks on us.
luma.com/po85yuqq
Most MCP traffic is internal, but the servers are exposed to from the public internet. Why? Speakeasy makes it trivially easy to move all your servers to your Tailnet. Put them on your private security of your AI workflows in seconds.
If you're in New York for @aiDotEngineer, come join. Drinks on us.
luma.com/po85yuqq
1
1
12
We're teaming up with @Tailscale to keep your MCP connections off the public internet.
Join us at a real speakeasy on the last day of AIE NYC, to chat about privacy in the age of AI.
Space is limited.
Join us at a real speakeasy on the last day of AIE NYC, to chat about privacy in the age of AI.
Space is limited.
1
1
14
We'll be at The @AIconference next week at Pier 48, San Francisco.
Come find us on the floor. We'll be talking through how teams put identity, policy, and an audit trail on every agent action.
We also have a shuffle board. See you there!
Come find us on the floor. We'll be talking through how teams put identity, policy, and an audit trail on every agent action.
We also have a shuffle board. See you there!
0
1
9
Today we're announcing our partnership with @googledevs, and open sourcing infrastructure for agent experience together ๐ค
Every enterprise is coming to the realization that developer experience is no longer the whole story. Agents are rapidly becoming the users integrating with need infrastructure built for them: Discoverable interfaces, structured output, context they can query at runtime.
Those are the foundations we've been building together.
Google is future-proofing the Gemini APIs' integration experience so agents get the same first-class treatment developers do, and Speakeasy is the platform generating and governing the artifacts that make that possible.
We've been building the Speakeasy AI Control Plane to provide enterprises the identity, policy enforcement, and observability over every agent connection. But before you implement control, you need robust connections.
We've been working with the Google team to Open Source the primitives needed for agents to build those connections.
The Speakeasy generator is now open source and includes everything you need to make your services consumable by agents: MCPs, CLIs, and SDKs
speakeasy.com/blog/partnerinโฆ
github.com/speakeasy-api/โฆ
Every enterprise is coming to the realization that developer experience is no longer the whole story. Agents are rapidly becoming the users integrating with need infrastructure built for them: Discoverable interfaces, structured output, context they can query at runtime.
Those are the foundations we've been building together.
Google is future-proofing the Gemini APIs' integration experience so agents get the same first-class treatment developers do, and Speakeasy is the platform generating and governing the artifacts that make that possible.
We've been building the Speakeasy AI Control Plane to provide enterprises the identity, policy enforcement, and observability over every agent connection. But before you implement control, you need robust connections.
We've been working with the Google team to Open Source the primitives needed for agents to build those connections.
The Speakeasy generator is now open source and includes everything you need to make your services consumable by agents: MCPs, CLIs, and SDKs
speakeasy.com/blog/partnerinโฆ
github.com/speakeasy-api/โฆ
4
2
11
Speakeasy is now available on the Microsoft commercial marketplace.
Entra ID and Intune were built to govern people, not agents.
We integrate with both to add policy enforcement, attribution, and an audit trail onto every agent action.
marketplace.microsoft.com/en-us/product/โฆ
Entra ID and Intune were built to govern people, not agents.
We integrate with both to add policy enforcement, attribution, and an audit trail onto every agent action.
marketplace.microsoft.com/en-us/product/โฆ
4
1
10
To close out August, we made the control plane easier to operate.
A shadow MCP block now opens into a reviewed approval, with the evidence assembled for the admin automatically.
The Platform MCP lets admins run the gateway, risk policies, and session porting from inside the
And session portability carries a live session, full context, from one agent to another.
Govern how your agents connect, secure, and operate.
A shadow MCP block now opens into a reviewed approval, with the evidence assembled for the admin automatically.
The Platform MCP lets admins run the gateway, risk policies, and session porting from inside the
And session portability carries a live session, full context, from one agent to another.
Govern how your agents connect, secure, and operate.
1
0
12
This week X was filled with discussions on being MCP-first or not. Whether or not its in app it's clear to providing MCP tools for agents is table stakes to ensuring customers are getting the most out of your product.
Speakeasy's MCP now lets your favorite agent manage your MCP ity policies. It comes with a guided setup in case you're new to MCP or the platform.
- Discover and provision governed endpoints for your 3rd party MCPs.
- Define a security policy in natural language: setup destructive tool call blocking, prompt injection, pii detection.
- Understand overall team usage and metrics.
The same MCP powers the in app assistant too.
Read more on the release speakeasy.com/blog/release-pโฆ
Speakeasy's MCP now lets your favorite agent manage your MCP ity policies. It comes with a guided setup in case you're new to MCP or the platform.
- Discover and provision governed endpoints for your 3rd party MCPs.
- Define a security policy in natural language: setup destructive tool call blocking, prompt injection, pii detection.
- Understand overall team usage and metrics.
The same MCP powers the in app assistant too.
Read more on the release speakeasy.com/blog/release-pโฆ
1
2
4
92% of enterprise CISOs say they can't fully identify their AI agents.
IdPs were built for people.
Today, agents run on borrowed identity.
Speakeasy gives them their own.
IdPs were built for people.
Today, agents run on borrowed identity.
Speakeasy gives them their own.
1
1
11
Our CEO Sagar Batchu appeared on @cisoseries's Security You Should Know podcast to tell CISOs that 'you should be able to adopt AI everywhere and still prove it's governed.'
youtube.com/watch?v=8jJ_kuโฆ
youtube.com/watch?v=8jJ_kuโฆ
0
0
10
@OpenRouter is the model gateway of choice for our AI Control Plane.
Two ways to run it: We'll create platform-provisioned keys with billing warnings at 50/75/90/100% of your monthly spend cap, or you can bring your own OpenRouter key and scope it per project and use case. bill to your OpenRouter account and show up in its activity view.
Build on our AI control plane with whatever model you want ๐
Two ways to run it: We'll create platform-provisioned keys with billing warnings at 50/75/90/100% of your monthly spend cap, or you can bring your own OpenRouter key and scope it per project and use case. bill to your OpenRouter account and show up in its activity view.
Build on our AI control plane with whatever model you want ๐
1
3
15
Speakeasy now integrates with LiteLLM.
If LiteLLM is your routing layer, Speakeasy plugs in as its enforcement point. Setup is a few environment variables and a guardrail fragment, with no custom callback code, and each instance declares whether to fail closed or open if tform is lost.
- ๐ฃ๐ผ๐น๐ถ๐ฐ๐ถ๐ฒ๐ ๐ฟ๐๐ป ๐ฏ๐ฒ๐ณ๐ผ๐ฟ๐ฒ ๐ถ๐ป๐ณ๐ฒ๐ฟ๐ฒ๐ป๐ฐ๐ฒ. Enforcement happens through LiteLLM's Generic Guardrail API, so a violating prompt is blocked before any provider processes it, and before you're charged for it.
- ๐ข๐ฏ๐๐ฒ๐ฟ๐๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ ๐ถ๐ป๐ฐ๐น๐๐ฑ๐ฒ๐ฑ. Model, tokens, cost, and duration flow into Observe and Costs over OpenTelemetry, with prompt content kept off the telemetry path.
- ๐ฅ๐ฒ๐ฎ๐น ๐๐๐ฒ๐ฟ ๐ฎ๐๐๐ฟ๐ถ๐ฏ๐๐๐ถ๐ผ๐ป. Virtual-key emails map to members of your org, so enforcement actions and spend attach to individuals instead of a shared key.
speakeasy.com/blog/release-lโฆ
If LiteLLM is your routing layer, Speakeasy plugs in as its enforcement point. Setup is a few environment variables and a guardrail fragment, with no custom callback code, and each instance declares whether to fail closed or open if tform is lost.
- ๐ฃ๐ผ๐น๐ถ๐ฐ๐ถ๐ฒ๐ ๐ฟ๐๐ป ๐ฏ๐ฒ๐ณ๐ผ๐ฟ๐ฒ ๐ถ๐ป๐ณ๐ฒ๐ฟ๐ฒ๐ป๐ฐ๐ฒ. Enforcement happens through LiteLLM's Generic Guardrail API, so a violating prompt is blocked before any provider processes it, and before you're charged for it.
- ๐ข๐ฏ๐๐ฒ๐ฟ๐๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ ๐ถ๐ป๐ฐ๐น๐๐ฑ๐ฒ๐ฑ. Model, tokens, cost, and duration flow into Observe and Costs over OpenTelemetry, with prompt content kept off the telemetry path.
- ๐ฅ๐ฒ๐ฎ๐น ๐๐๐ฒ๐ฟ ๐ฎ๐๐๐ฟ๐ถ๐ฏ๐๐๐ถ๐ผ๐ป. Virtual-key emails map to members of your org, so enforcement actions and spend attach to individuals instead of a shared key.
speakeasy.com/blog/release-lโฆ
1
0
9
We've joined the @AgenticAIFdn ๐ค
The AAIF is the home for the open standards that agents increasingly rely on and is leading the charge on several topics that are near and dear to our hearts: Security, safety, trust and governance.
Our AI control plane sits between AI and the o be effective. We enforce identity-scoped access, inspect traffic for threats and produce audit trails for compliance.
We're excited to contribute the insights we've gleaned back to the working groups that are shaping the future of Agent standards.
The AAIF is the home for the open standards that agents increasingly rely on and is leading the charge on several topics that are near and dear to our hearts: Security, safety, trust and governance.
Our AI control plane sits between AI and the o be effective. We enforce identity-scoped access, inspect traffic for threats and produce audit trails for compliance.
We're excited to contribute the insights we've gleaned back to the working groups that are shaping the future of Agent standards.
1
2
11
Thank you @moov for having us at another great #fintechdevcon!
Really encouraging to hear how top-of-mind AI governance was this year.
Really encouraging to hear how top-of-mind AI governance was this year.
0
0
9
3 updates to the Speakeasy AI Control Plane from July:
๐ฆ๐ธ๐ถ๐น๐น๐ ๐ ๐ฎ๐ป๐ฎ๐ด๐ฒ๐บ๐ฒ๐ป๐: a system of record for every agent skill, versioned and validated, with usage and performance analytics.
๐๐ด๐ฒ๐ป๐ ๐๐๐ฑ๐ด๐ฒ๐๐: one enforceable budget per person, across every agent ๐๐ถ๐ฐ๐ฒ๐ป๐๐ฒ ๐ข๐ฏ๐๐ฒ๐ฟ๐๐ฎ๐ฏ๐ถ๐น๐ถ๐๐: personal and enterprise AI usage tracked in one place, at the device level.
Govern every agent, from one place.
๐ฆ๐ธ๐ถ๐น๐น๐ ๐ ๐ฎ๐ป๐ฎ๐ด๐ฒ๐บ๐ฒ๐ป๐: a system of record for every agent skill, versioned and validated, with usage and performance analytics.
๐๐ด๐ฒ๐ป๐ ๐๐๐ฑ๐ด๐ฒ๐๐: one enforceable budget per person, across every agent ๐๐ถ๐ฐ๐ฒ๐ป๐๐ฒ ๐ข๐ฏ๐๐ฒ๐ฟ๐๐ฎ๐ฏ๐ถ๐น๐ถ๐๐: personal and enterprise AI usage tracked in one place, at the device level.
Govern every agent, from one place.
1
2
12
Huge thanks to our friends at @moov for hosting #fintechdevcon! We can't wait to be there next month, talking AI governance with the best and brightest in fintech.
And if that sounds like you, stay tuned for our newest case study on how one of the world's leading crypto payments its company-wide AI usage with Speakeasy.
And if that sounds like you, stay tuned for our newest case study on how one of the world's leading crypto payments its company-wide AI usage with Speakeasy.
0
0
11
Shed some light on shadow MCPs...
You can't govern an MCP server you don't know exists. Most companies have more of them than they think, spun up by whoever needed one that afternoon.
Shadow MCP detection finds the servers nobody registered, so the unsanctioned ones land on the ved ones. From there you can review them, pin them, or shut them down.
How it works: speakeasy.com/blog/release-sโฆ
You can't govern an MCP server you don't know exists. Most companies have more of them than they think, spun up by whoever needed one that afternoon.
Shadow MCP detection finds the servers nobody registered, so the unsanctioned ones land on the ved ones. From there you can review them, pin them, or shut them down.
How it works: speakeasy.com/blog/release-sโฆ
0
0
1
Legacy IAM runs on one assumption.
Once it confirms a session belongs to a person you trust, it trusts that person to use their permissions responsibly.
Agents break that assumption.
Within a verified session, an agent can act on any instruction it finds, at machine speed.
s to work differently. Speakeasy ties every agent and every tool call back to a real person and their permissions.
How it works: speakeasy.com/blog/release-aโฆ
Once it confirms a session belongs to a person you trust, it trusts that person to use their permissions responsibly.
Agents break that assumption.
Within a verified session, an agent can act on any instruction it finds, at machine speed.
s to work differently. Speakeasy ties every agent and every tool call back to a real person and their permissions.
How it works: speakeasy.com/blog/release-aโฆ
0
1
10
Half your Claude usage probably isn't on your team plan.
Engineers hit the limits on their enterprise seats fast, so they run personal Claude Max or Pro accounts on the same laptop for the headroom.
It's cheaper per token, but it doesn't show up on your enterprise dashboards.
entire AI spend across every provider and every account your team is using.
Run Max where it's cheaper, keep enterprise where you need it, and govern both the same way.
How it works: speakeasy.com/blog/release-pโฆ
Engineers hit the limits on their enterprise seats fast, so they run personal Claude Max or Pro accounts on the same laptop for the headroom.
It's cheaper per token, but it doesn't show up on your enterprise dashboards.
entire AI spend across every provider and every account your team is using.
Run Max where it's cheaper, keep enterprise where you need it, and govern both the same way.
How it works: speakeasy.com/blog/release-pโฆ
0
1
11
An agent can wipe a production table with sufficient permissions and one bad instruction, whether careless or malicious.
Speakeasy risk policies put guardrails on every prompt and tool call: prompt-injection detection, destructive-command flagging, and plain-language rules e. Each policy is scoped to an audience, with clear block messages and per-case exceptions.
How they work: speakeasy.com/blog/release-aโฆ
Speakeasy risk policies put guardrails on every prompt and tool call: prompt-injection detection, destructive-command flagging, and plain-language rules e. Each policy is scoped to an audience, with clear block messages and per-case exceptions.
How they work: speakeasy.com/blog/release-aโฆ
0
0
8
@verrsane @sagar_batchu This is neat and makes so much sense for Glean to provide integrated into the existing context layer. Here's one of ours :)
1
0
1
In June we widened the Speakeasy AI control plane beyond session security, with four new ways to govern how your agents connect, spend, and act.
โข MCP Tunnels: govern private MCP servers without exposing them โ speakeasy.com/blog/release-mโฆ
โข Cost attribution: see agent spend by MCP server โ speakeasy.com/resources/ai-cโฆ
โข Prompt-based policies: write a risk policy in plain language โ speakeasy.com/blog/release-aโฆ
โข Managed auth: one dashboard for every issuer, client, and session โ speakeasy.com/blog/release-aโฆ
โข MCP Tunnels: govern private MCP servers without exposing them โ speakeasy.com/blog/release-mโฆ
โข Cost attribution: see agent spend by MCP server โ speakeasy.com/resources/ai-cโฆ
โข Prompt-based policies: write a risk policy in plain language โ speakeasy.com/blog/release-aโฆ
โข Managed auth: one dashboard for every issuer, client, and session โ speakeasy.com/blog/release-aโฆ
0
0
7
Blocking MCP servers doesn't stop shadow AI. It just moves it to personal laptops.
The fix is to make the sanctioned path the easy one. With MCP authorization you can define roles scoped to specific people, teams, and MCP servers. Paired with the Speakeasy registry, IT can stand ry of MCP servers for the whole company and trust it's configured correctly for everyone who connects.
Roles are built from granular scopes, so an order-processing role can reach one server and nothing else, and they sync from your identity provider so access follows the org chart instead of a spreadsheet.
We wrote up how it works, and why authorization has to be enforced at the gateway, not just defined in your identity provider: speakeasy.com/blog/release-mโฆ
The fix is to make the sanctioned path the easy one. With MCP authorization you can define roles scoped to specific people, teams, and MCP servers. Paired with the Speakeasy registry, IT can stand ry of MCP servers for the whole company and trust it's configured correctly for everyone who connects.
Roles are built from granular scopes, so an order-processing role can reach one server and nothing else, and they sync from your identity provider so access follows the org chart instead of a spreadsheet.
We wrote up how it works, and why authorization has to be enforced at the gateway, not just defined in your identity provider: speakeasy.com/blog/release-mโฆ
0
1
9
You shouldn't have to wait for the invoice to find out what your AI cost.
Speakeasy sits in the path of every AI call, so spend is managed in real time, not reconciled after the fact. Personal licenses, enterprise seats, provider APIs, and your model router report into one ed with the team, user, agent, and model behind it.
Set budgets per team or agent, get alerted before the line, and cap a runaway agent before it drains the budget.
One view of every dollar of AI spend: speakeasy.com/use-cases/ai-cโฆ
Speakeasy sits in the path of every AI call, so spend is managed in real time, not reconciled after the fact. Personal licenses, enterprise seats, provider APIs, and your model router report into one ed with the team, user, agent, and model behind it.
Set budgets per team or agent, get alerted before the line, and cap a runaway agent before it drains the budget.
One view of every dollar of AI spend: speakeasy.com/use-cases/ai-cโฆ
0
2
9
Cyberhaven's AI Adoption & Risk Report found that 40% of enterprise AI interactions involve sensitive data. The average employee feeds proprietary information into an AI tool once every three days.
Your existing DLP watches email, browsers, and endpoints.
The places people leak datathere.
When an agent pulls a customer record to answer a question and passes it to an outside model, or repeats an API key it picked up in a tool response, the data moves machine to machine, inside prompts and tool calls your DLP was never able to read.
The Speakeasy AI Control Plane inspects every message between your users, models, and tools for sensitive data, from secrets and PII to financial and healthcare records.
Detection is context-aware, combining deterministic libraries with ML instead of matching patterns, and you can add your own rules with a regex or a plain-language description.
When something matches, flag it for audit or block it before it leaves the session.
Your existing DLP watches email, browsers, and endpoints.
The places people leak datathere.
When an agent pulls a customer record to answer a question and passes it to an outside model, or repeats an API key it picked up in a tool response, the data moves machine to machine, inside prompts and tool calls your DLP was never able to read.
The Speakeasy AI Control Plane inspects every message between your users, models, and tools for sensitive data, from secrets and PII to financial and healthcare records.
Detection is context-aware, combining deterministic libraries with ML instead of matching patterns, and you can add your own rules with a regex or a plain-language description.
When something matches, flag it for audit or block it before it leaves the session.
1
1
7
Ask most engineering leaders how many AI agents are running across their org and you'll get a shrug.
Not because they don't care. Because there's genuinely no way to know. Teams spin up MCP servers, connect agents to internal tools, experiment with models. None of it goes ol point.
This was a "we'll figure it out later" problem. Except regulators just gave everyone a deadline. The EU AI Act requires companies to prove governance, traceability, and oversight across AI systems classified as high-risk. And the bar for what counts as "high-risk" is wider than most people think.
The companies scrambling to comply when the deadline hits are the same ones who said "we'll add observability later" about their APIs in 2018.
We know how that worked out.
Read more on the EU AI Act's requirements and deadlines here: speakeasy.com/blog/eu-ai-act
Not because they don't care. Because there's genuinely no way to know. Teams spin up MCP servers, connect agents to internal tools, experiment with models. None of it goes ol point.
This was a "we'll figure it out later" problem. Except regulators just gave everyone a deadline. The EU AI Act requires companies to prove governance, traceability, and oversight across AI systems classified as high-risk. And the bar for what counts as "high-risk" is wider than most people think.
The companies scrambling to comply when the deadline hits are the same ones who said "we'll add observability later" about their APIs in 2018.
We know how that worked out.
Read more on the EU AI Act's requirements and deadlines here: speakeasy.com/blog/eu-ai-act
0
0
5
Before you roll agents out across a company, you have to be able to answer three questions: what is the agent reading, what is it connected to, and what data can it move.
This month we shipped the features that answer them, expanding the security layer of the Speakeasy AI ๐ ๐๐ป๐ท๐ฒ๐ฐ๐๐ถ๐ผ๐ป ๐๐ฒ๐๐ฒ๐ฐ๐๐ถ๐ผ๐ป: catch malicious instructions hidden in anything your agent reads, before they execute
โ ๐ฆ๐ต๐ฎ๐ฑ๐ผ๐ ๐ ๐๐ฃ ๐๐ฒ๐๐ฒ๐ฐ๐๐ถ๐ผ๐ป: every MCP server mapped to its real endpoint and checked against your approved list at session start
โ ๐ ๐๐ฃ ๐ฆ๐ฒ๐ฟ๐๐ฒ๐ฟ ๐๐ฝ๐ฝ๐ฟ๐ผ๐๐ฎ๐น ๐ช๐ผ๐ฟ๐ธ๐ณ๐น๐ผ๐๐: one-click access requests, approve org-wide or project-wide, policies inherited automatically
โ ๐๐ฎ๐๐ฎ ๐๐ผ๐๐ ๐ฃ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ถ๐ผ๐ป: detect secrets, PII, financial, and healthcare data, and flag or block it before it leaves the session
Every check runs in real time, on every message in and out of your agents. More to come.
This month we shipped the features that answer them, expanding the security layer of the Speakeasy AI ๐ ๐๐ป๐ท๐ฒ๐ฐ๐๐ถ๐ผ๐ป ๐๐ฒ๐๐ฒ๐ฐ๐๐ถ๐ผ๐ป: catch malicious instructions hidden in anything your agent reads, before they execute
โ ๐ฆ๐ต๐ฎ๐ฑ๐ผ๐ ๐ ๐๐ฃ ๐๐ฒ๐๐ฒ๐ฐ๐๐ถ๐ผ๐ป: every MCP server mapped to its real endpoint and checked against your approved list at session start
โ ๐ ๐๐ฃ ๐ฆ๐ฒ๐ฟ๐๐ฒ๐ฟ ๐๐ฝ๐ฝ๐ฟ๐ผ๐๐ฎ๐น ๐ช๐ผ๐ฟ๐ธ๐ณ๐น๐ผ๐๐: one-click access requests, approve org-wide or project-wide, policies inherited automatically
โ ๐๐ฎ๐๐ฎ ๐๐ผ๐๐ ๐ฃ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ถ๐ผ๐ป: detect secrets, PII, financial, and healthcare data, and flag or block it before it leaves the session
Every check runs in real time, on every message in and out of your agents. More to come.
0
1
8
1/ AI assistants today are solo-developer runtimes wearing an enterprise badge.
Persistent hosts. Native shells. Every inbound message dropped straight into the model's context.
Wonderful for indie hackers. A breach waiting to happen for a 10,000-person enterprise.
Persistent hosts. Native shells. Every inbound message dropped straight into the model's context.
Wonderful for indie hackers. A breach waiting to happen for a 10,000-person enterprise.
4
0
1
PlanetScale is a Speakeasy customer ๐ค
They don't just build a great database, they put in the work to build a great platform! Whether a developer is prompting an AI agent or writing infrastructure-as-code, PlanetScale just works.
We've loved collaborating with them to launch their Terraform provider and MCP server and look forward to many more years of partnership!
They don't just build a great database, they put in the work to build a great platform! Whether a developer is prompting an AI agent or writing infrastructure-as-code, PlanetScale just works.
We've loved collaborating with them to launch their Terraform provider and MCP server and look forward to many more years of partnership!
1
2
14
Most companies think AI governance is a policy problem.
It is. But it's also an architecture problem, a complicated one...
Here's what a proper AI control plane actually looks like, and why each layer matters:
Every AI interaction flows through 3 zones:
01 ยท Callers (who's eers, sales, finance, ops
โ Machine callers: Cursor, Copilot, Devin, custom agents
โ Autonomous workers: role-based agents like OpenClaw running without a human in the loop
02 ยท Control plane (the governed path)
Identity & access: spans every call. SSO, role/team scoping, credential management via OIDC, SAML, SCIM. Not "who set up their personal account" but "who is this person in the org and what are they entitled to."
LLM gateway: sits in front of every model call. Multi-provider routing, rate limiting, caching. One lane for every prompt.
MCP gateway: sits in front of every tool call. Tool registry and routing, per-team scoping. One lane for every action.
Policy & threat: applied to every call. Executable allow/deny rules. Real-time detection of PII, data exfiltration, prompt injection, shadow tools. The policy that lives in a Confluence doc is not a control. This is.
Observability & audit: spans every call. Every prompt, response, and tool call captured. Adoption analytics. SIEM integration. OpenTelemetry-native.
03 ยท Destinations (what they reach)
โ LLMs: Claude, GPT, Gemini, internal models
โ Enterprise systems: SaaS apps, internal APIs, databases, data warehouses
Enterprises need to assemble a control plane that spans ALL traffic, not just some of it.
An LLM gateway that sees prompts but not tool calls isn't a control plane. An identity layer that covers Claude but not Cursor isn't a control plane. Governance only works when it's a property of the path itself, not bolted on per-tool.
That's what we're solving at Speakeasy
Link to the full reference architecture in the comments ๐
It is. But it's also an architecture problem, a complicated one...
Here's what a proper AI control plane actually looks like, and why each layer matters:
Every AI interaction flows through 3 zones:
01 ยท Callers (who's eers, sales, finance, ops
โ Machine callers: Cursor, Copilot, Devin, custom agents
โ Autonomous workers: role-based agents like OpenClaw running without a human in the loop
02 ยท Control plane (the governed path)
Identity & access: spans every call. SSO, role/team scoping, credential management via OIDC, SAML, SCIM. Not "who set up their personal account" but "who is this person in the org and what are they entitled to."
LLM gateway: sits in front of every model call. Multi-provider routing, rate limiting, caching. One lane for every prompt.
MCP gateway: sits in front of every tool call. Tool registry and routing, per-team scoping. One lane for every action.
Policy & threat: applied to every call. Executable allow/deny rules. Real-time detection of PII, data exfiltration, prompt injection, shadow tools. The policy that lives in a Confluence doc is not a control. This is.
Observability & audit: spans every call. Every prompt, response, and tool call captured. Adoption analytics. SIEM integration. OpenTelemetry-native.
03 ยท Destinations (what they reach)
โ LLMs: Claude, GPT, Gemini, internal models
โ Enterprise systems: SaaS apps, internal APIs, databases, data warehouses
Enterprises need to assemble a control plane that spans ALL traffic, not just some of it.
An LLM gateway that sees prompts but not tool calls isn't a control plane. An identity layer that covers Claude but not Cursor isn't a control plane. Governance only works when it's a property of the path itself, not bolted on per-tool.
That's what we're solving at Speakeasy
Link to the full reference architecture in the comments ๐
0
0
1
One platform to connect, secure, control, and observe every AI agent across your org.
This is the AI control plane.
This is the AI control plane.
5
1
6
The top MCP question from enterprise leaders right now:
"How do I curate an approved list of servers for my org?"
With 13,000+ MCP servers out there and new ones shipping daily, "just let people install things" isn't a strategy.
So we teamed up with @tadasayy and pulsemcp to build a curated catalog of trusted MCP servers purpose-built for Speakeasy customers rolling out MCP at scale.
Production-grade, vetted, enterprise-ready. ๐ค
"How do I curate an approved list of servers for my org?"
With 13,000+ MCP servers out there and new ones shipping daily, "just let people install things" isn't a strategy.
So we teamed up with @tadasayy and pulsemcp to build a curated catalog of trusted MCP servers purpose-built for Speakeasy customers rolling out MCP at scale.
Production-grade, vetted, enterprise-ready. ๐ค
0
3
9
MCP authorization is live.
Define roles. Scope access to specific MCP servers. Plug into your existing SSO and SCIM provisioning.
One registry. Every team gets exactly the servers they need. Nobody gets what they shouldn't.
Define roles. Scope access to specific MCP servers. Plug into your existing SSO and SCIM provisioning.
One registry. Every team gets exactly the servers they need. Nobody gets what they shouldn't.
1
0
4
Dennis Babyak just joined Speakeasy to help build our AI control plane.
He's coming from Bird with deep experience building systems at scale. Exactly the kind of engineer you want working on infrastructure that gives teams real visibility and control over their AI integrations.
Welcome, Dennis ๐ค
He's coming from Bird with deep experience building systems at scale. Exactly the kind of engineer you want working on infrastructure that gives teams real visibility and control over their AI integrations.
Welcome, Dennis ๐ค
0
1
4
Autumn built an open-source billing layer that saves AI teams weeks of Stripe integration. 3 functions. No webhooks. Usage tracking that just works.
Now they're using Speakeasy to generate SDKs that match the quality of their API.
Welcome aboard, @autumnpricing ๐
Now they're using Speakeasy to generate SDKs that match the quality of their API.
Welcome aboard, @autumnpricing ๐
1
1
11
We're excited to welcome Alex Martin to the Speakeasy engineering team! Alex joins us from Render, where he was the first UX engineering hire. He's joining us to repeat the feat, by bringing his design and product chops to our SF office. Excited to make the pixels dance! ๐บ
0
1
8
Bienvenue Qynn! ๐
Qynn has been an incredible contributor as a contractor for the past couple of years, and we couldn't be more excited to have them join us permanently. Their deep knowledge of our platform and the work they've already done have made this a long time coming.
ynn we're lucky to have you!
Qynn has been an incredible contributor as a contractor for the past couple of years, and we couldn't be more excited to have them join us permanently. Their deep knowledge of our platform and the work they've already done have made this a long time coming.
ynn we're lucky to have you!
0
0
8
@boristane @ojusave @render @jamesperkins @unkeydev @ndimares Full playlist is here: youtube.com/watch?v=9ZwwqAโฆ
0
0
1
@boristane @ojusave @render @jamesperkins @unkeydev @ndimares building an AI-powered slackbot on Speakeasy MCP platform
1
0
1
We hosted Drinks & Demos in SF with our friends at Unkey and Render. For anyone who couldn't make it, the recordings are live now. Want a little preview? Here are some of @boristane's thoughts on the future of the SDLC ๐ฅ
1
0
6
Last week the Speakeasy team got together in Madrid ๐ช๐ธ
We operate out of our SF & London offices, so offsites are one of the few times both teams are all in the same room. The energy when the whole team gets together in person is special every time.
Big things coming ๐
We operate out of our SF & London offices, so offsites are one of the few times both teams are all in the same room. The energy when the whole team gets together in person is special every time.
Big things coming ๐
0
0
1
Security is the #1 reason MCP rollouts stall at the enterprise level.
Not because teams can't build secure servers. Because every server becomes its own auth conversation. ๐งต
Not because teams can't build secure servers. Because every server becomes its own auth conversation. ๐งต
1
0
1
Huge thank you to everyone who came out to Demos & Drinks last night!
We got together with SF's AI developers and product builders for a night of real demos and conversations with our friends at @unkeydev and @render.
Shoutout to our demo lineup:
๐น @boristane showing off OC at hve building on Render's new MCP server
๐น @jamesperkins showing off Unkey deploy
๐น @ndimares running an MCP-enabled AI Slackbot on Speakeasy
Four demos, four builders, and a room full of people who'd rather ship than speculate. The energy in the room was a reminder that the best way to move this space forward is to just show what's possible.
If you missed this one, stay tuned for the next one.
We got together with SF's AI developers and product builders for a night of real demos and conversations with our friends at @unkeydev and @render.
Shoutout to our demo lineup:
๐น @boristane showing off OC at hve building on Render's new MCP server
๐น @jamesperkins showing off Unkey deploy
๐น @ndimares running an MCP-enabled AI Slackbot on Speakeasy
Four demos, four builders, and a room full of people who'd rather ship than speculate. The energy in the room was a reminder that the best way to move this space forward is to just show what's possible.
If you missed this one, stay tuned for the next one.
0
3
15