Observe · Agent telemetry

Every agent action,accounted for.

Speakeasy sits in the path of every agent interaction. Security and engineering leadership get one audit trail of every session, tool call, token, and dollar, attributed to a named person, in real time.

Talk to usRead the docs

Audit trail · All agents

Live
nadia@acme · github.create_pull_request$0.04
support-bot · zendesk.reply$0.02
AWS key detected in promptblocked
milo@acme · cursor · agent run$0.09
growth · 92% of monthly budgetalert

44,120

Tool calls

$1,284.61

Spend today

12

Flagged

Trusted byVerizonMistralCloudinaryLaunchDarklyFivetranPlanetScale

01 · The problem

You can't govern what you can't see

Without observability
What are agents touching?
Agents reach repos, CRMs, and internal APIs through personal accounts and pasted keys. No record of what data moved, or where
Who answers the audit?
When security or compliance asks what an agent did last quarter, the evidence is scattered across provider logs that were never designed for it
What does it cost?
Token spend climbs across teams with no attribution. Costs surface on the invoice, long after the runs that drove them
On the control plane
What are agents touching?
Every session, tool call, and permission change in one searchable audit trail, attributed to a named person
Who answers the audit?
One record of who asked what, when, against which system, exportable to your SIEM
What does it cost?
Every token attributed to a team and budget, enforced in real time

Speakeasy's AI control plane has been indispensable in enabling Fivetran's AI transformation.

Eli Davis

Fivetran

Fivetran

02 · The control plane

One audit trail for every agent

Speakeasy sits between your agents and every system they touch. Observe every interaction, secure it in flight, control spend and access, and report with evidence, from one platform.

Observes every agentAnthropicAnthropicOpenAIOpenAIGoogleGoogleCursorCursorGitHub CopilotCopilotProvider APIsModel routersCustom agents

03 · Why now

Adoption is outrunning oversight

Cloud taught the enterprise to demand visibility before scale. Agents are climbing the same curve faster, acting on production systems and real data. Boards are asking what agents touch and what the spend returns, and provider logs cannot answer either question.

Read the Fivetran story

$644B

AI spending in 2025

Gartner forecast, 2025.

78%

Employees bringing their own AI to work

Microsoft Work Trend Index, 2024.

98%

FinOps teams now managing AI spend

Up from 31% in 2024. FinOps Foundation, 2026.

Every agent

One view of every agent

Teams rarely settle on a single agent. The same audit trail, guardrails, and budget controls apply across Anthropic, OpenAI, Google, and the agents your teams build themselves.

AnthropicOpenAIGoogleCursor
GitHub Copilot
OpenCode

Questions

What does agent observability cover?
Every session, tool call, prompt, and response that crosses the control plane, plus the tokens and dollars behind them. Each event is attributed to a named person, team, and agent, and the full history is searchable from one place.
Can we attribute agent activity to specific people and teams?
Yes. Speakeasy plugs into your identity provider, so every session maps to a named employee and their team. Activity and spend roll up by person, team, agent, and model, ready for review or chargeback.
How does this help with security and compliance?
Every interaction is inspected in real time: PII, secrets, and source code are blocked before they leave the org, and prompt injection and shadow tools are flagged. The audit trail exports to your SIEM, and the platform is SOC 2 Type II and ISO 27001 certified.
Does observability include cost?
Yes. Every call is metered at the token level and priced, so spend sits in the same trail as activity. Budgets per team, agent, or key are enforced in real time, with alerts before the threshold and a hard cap at it.
Can we see usage from employees' personal AI subscriptions?
Yes. Most teams have employees running personal Claude, ChatGPT, and Cursor plans alongside company seats, and that shadow usage is a real blind spot. Speakeasy attributes activity from personal and enterprise licenses to the same person, so the full footprint is visible in one place.
What happens when an agent runs away?
Speakeasy detects abnormal token bursts and loops, then caps the agent before it drains the budget. Alerts fire in real time so the team can step in.
Can we export the data into our own tools?
Yes. The audit trail and cost records export in standard formats to SIEMs, BI tools, data warehouses, and finance systems.
Does this change how people use their agents?
No. Speakeasy layers onto Claude, Codex, Cursor, and the rest without changing the client. People keep working the way they already do; the control plane records, inspects, and enforces behind the scenes.
How do we get started?
Book a demo and our team will walk through connecting your agents, wiring your identity provider, and configuring policies and budgets for your organization.

AI everywhere.

Control here.

Talk to usRead the docs