What do I see in Mission Control?

Six live KPIs: Requests, Avg Latency, Vinkius Reliability, Tokens, DLP Protected, and Cost Saved. Each with a health indicator and trend, plus an AI Briefing that summarizes what needs your attention, and a Live Activity table of the latest tool executions across all MCP servers.

How does failure attribution work?

Every error is assigned to the party that caused it. Agent errors mean the caller sent a bad request, Upstream errors mean the MCP provider failed, and Vinkius errors mean we failed or blocked the call ourselves. The Reliability KPI excludes Vinkius errors, so we never inflate our own uptime.

Can I see how much latency Vinkius adds?

Yes. Every request is decomposed into upstream API latency and Vinkius governance overhead, shown side by side in the Live Activity table and averaged per server in Server Traffic. Governance overhead is typically tens of milliseconds.

How does DLP Protection shield my data?

You define redaction patterns like *.email or *.credit_card that apply to every conversation in the workspace. Matching fields are masked before the AI model or the connector sees them, and each redaction is counted in the Security Posture reports.

What does the FinOps Guard actually do?

It truncates array responses after a configurable number of items, optionally compresses tool descriptions, and attributes exact token cost per execution against your cost per million tokens rate, turning AI Spend into a precise, per Connector ledger.

What happens when the Circuit Breaker trips?

When request rates exceed your budget ceiling inside the detection window, agent traffic is halted automatically and your budget is protected. The circuit auto resets after the cooldown period, or you can approve resumption manually at any time.

Is there an emergency stop for all agents?

Yes. The Global Emergency Halt deactivates every active Connector, revokes all tokens, and terminates all running sessions across your account. It requires typing HALT ALL to confirm, and connectors can be restored afterwards. Revoked tokens must be issued again.

Providers · PromptApps · Response

PromptSummarize my unread Slack messages

PromptList my open pull requests

PromptFind my Q3 roadmap page

PromptCreate a bug from this stack trace

PromptDraft a reply to the Stripe invoice

PromptWhich sprint items are blocked?

PromptShow deals closing this month

PromptAny failed payments today?

PromptSummarize my unread Slack messages

PromptList my open pull requests

PromptFind my Q3 roadmap page

PromptCreate a bug from this stack trace

PromptDraft a reply to the Stripe invoice

PromptWhich sprint items are blocked?

PromptShow deals closing this month

PromptAny failed payments today?

ResponseBug ticket created in the sprint board

ResponseQ3 Roadmap · edited 2h ago

ResponseOpen PRs retrieved for 3 repos

ResponseDigest ready for 4 channels

ResponseFailed charges flagged and retried

ResponsePipeline summary ready to review

ResponseBlocked items flagged for standup

ResponseDraft saved to Gmail

ResponseBug ticket created in the sprint board

ResponseQ3 Roadmap · edited 2h ago

ResponseOpen PRs retrieved for 3 repos

ResponseDigest ready for 4 channels

ResponseFailed charges flagged and retried

ResponsePipeline summary ready to review

ResponseBlocked items flagged for standup

ResponseDraft saved to Gmail

See everything your AI agents are doing. In real time.

The control plane for every Connector your agents use: live activity, latency attribution, failure responsibility, data shielding, cost guardrails, and an emergency halt. All in one dashboard.

Ask AI about AI Governance

One Control Plane

Twelve surfaces. Every question about your AI traffic, answered.

Eight report surfaces tell you what is happening. Four policy surfaces decide what is allowed to happen, and every one of them is one click away in the real dashboard.

Reports08

01Mission Control
02Agent Activity
03Server Traffic
04Access Tokens
05AI Spend
06Tool Reliability
07Security Posture
08Request Failures

04Settings04

Connector PolicyRate limits per Connector01
DLP ProtectionRedact *.email, *.credit_card02
FinOps GuardTruncate responses, cap spend03
Circuit BreakerAuto-halt at budget ceiling04

Mission Control

Six live KPIs, one glance, zero guessing.

Requests, latency, reliability, tokens, data protection, and cost savings. Each with its health indicator and trend, updating synchronously with every tool execution your agents make.

AI Briefing

Upgrade to unlock

Know exactly what needs your attention before you even look at the charts. One briefing. The full picture. The next move.

Requests

Total in this period

Avg Latency

API 88ms · Overhead 54ms

Vinkius Reliability

Agent 280 · Upstream 47 · Vinkius 15

Tokens

AI tokens processed

DLP Protected

Sensitive data redacted

Cost Saved

FinOps truncations applied

AI Agents Activities

last 30 days

30d agotoday

Request Volume & Latency

requests latency

Emergency Control

One switch stops every agent. Instantly.

When something goes wrong, you do not file a ticket or revoke keys one by one. You halt everything. Every Connector, every token, every session, in one synchronous action, gated behind an explicit typed confirmation.

Every active Connector is deactivated

Every access token is revoked

Every running session is terminated

Connectors can be restored afterwards; revoked tokens must be issued again. Protection you can undo, damage you can prevent.

Global Emergency Halt

armed

All active Connectors will be deactivated, all tokens will be revoked, and all running sessions will be terminated across your account. AI agents will immediately lose access to all Connectors. Use only in an emergency.

12 → 0 active
Connectors: 12 revoked
Tokens: 7 terminated
Sessions

Type HALT ALL to confirm

Live Activity

Every tool execution, attributed and explained.

Each call is logged with its server, tool, semantic action, token, outcome, and a full latency breakdown. Green means the upstream answered. Violet means Vinkius policy acted in flight. Amber means the caller erred. Red means the provider failed. You always know who is responsible.

MCP Server	Tool	Action	Token	Outcome	API	Overhead	Total	When
GitHub	`list_issues`	Query	Token 1	Enforced200	96ms	38ms	134ms	n/a
Jira	`create_task`	Mutation	Token 2	Enforced200	88ms	41ms	129ms	n/a
Notion	`search_docs`	Query	Token 3	Agent Err401	12ms	3ms	15ms	n/a
Slack	`send_message`	Mutation	Token 4	Upstream Err503	203ms	9ms	212ms	n/a
HubSpot	`get_user`	Destructive	Token 5	Vinkius Err500	n/a	n/a	8ms	n/a
Supabase	`fetch_resource`	Query	Token 6	OK200	74ms	22ms	96ms	n/a

Live Activity. Latest tool executions across all MCP servers

Attribution

API latency versus Vinkius overhead. Failures by responsible party.

The only observability layer that separates what the provider took from what governance added, and splits every failure into Agent, Upstream, or Vinkius.

Latency split

Total response time decomposed into upstream API time and Vinkius governance overhead, on every request.

GET /repos/vinkius-labs/mcpfusion0ms

API 88msOverhead 54ms

Failure breakdown

Errors are never a mystery: each one is assigned to the party that caused it.

Agent Err: The caller sent a bad request: bad input, expired token, missing scope.
Upstream Err: The MCP provider failed: rate limits, outages, upstream 5xx.
Vinkius Err: We failed or blocked the call ourselves, for example via SSRF protection.

Tool Health Matrix

Latency × Failure rate · bubble = call volume

Policy Surfaces

Controls that act in flight, not after the fact.

Four policy surfaces govern every request before it reaches your budget, your data, or your rate limits. These are the real editors your team uses.

DLP Protection

Global Data Shielding

Default redaction patterns applied to every conversation in the workspace. Sensitive fields never reach the AI model.

*.email*.password*.secret*.credit_card*.card_number*.ssn*.phone*.api_key*.token*.date_of_birth*.ip_address*.bank_account*.iban

*.credit_card, users[*].ssn…

* matches any object. *.email protects every email field, items[*].credit_card protects array items.

FinOps Guard

Cost Guard

Apply response and payload controls before requests reach your budget limits.

Max Array Items0 items

Truncates array responses after this many items

Toon Compression

Compress tool descriptions to reduce token overhead from schema definitions.

Cost attribution rate

Estimated cost per 1M tokens (USD), used to calculate spending in analytics.

3.00

Circuit Breaker

Rate Protection

Automatically halts agent traffic when request rates exceed your defined budget ceiling.

5,000
Max Requests: 5 min
Detection Window: 15 min
Cooldown Period

Circuit Breaker Tripped

Agent traffic is currently halted. Your budget is protected. Approve resumption when ready. Auto resets in 12m 41s.

Connector Policy

How Connectors expose tools and behave across your account.

Tool exposition Flat

Each endpoint becomes an individual AI tool. Best for small APIs (< 20 tools).

Grouped

Related endpoints share one AI tool with an action discriminator. Reduces token usage.

Deployment approval

Require a secondary approval before deployments are activated in production. Implements the four eyes principle for high risk AI systems (EU AI Act Art. 14(5)).

AI Spend

Know the exact cost of every tool call.

Token expenditure attributed per Connector against your configured LLM rate, with FinOps savings measured where truncation and compression reduced the payload.

Estimated Cost

$8.52

2,840,000 tokens consumed

FinOps Savings

$1.86

620,000 tokens saved

Cost per Request

$0.002

4,230 requests

FinOps ROI

21.8%

savings vs spend

FAQ

AI Governance on Vinkius.

01
What do I see in Mission Control?
Six live KPIs: Requests, Avg Latency, Vinkius Reliability, Tokens, DLP Protected, and Cost Saved. Each with a health indicator and trend, plus an AI Briefing that summarizes what needs your attention, and a Live Activity table of the latest tool executions across all MCP servers.
02
How does failure attribution work?
Every error is assigned to the party that caused it. Agent errors mean the caller sent a bad request, Upstream errors mean the MCP provider failed, and Vinkius errors mean we failed or blocked the call ourselves. The Reliability KPI excludes Vinkius errors, so we never inflate our own uptime.
03
Can I see how much latency Vinkius adds?
Yes. Every request is decomposed into upstream API latency and Vinkius governance overhead, shown side by side in the Live Activity table and averaged per server in Server Traffic. Governance overhead is typically tens of milliseconds.
04
How does DLP Protection shield my data?
You define redaction patterns like *.email or *.credit_card that apply to every conversation in the workspace. Matching fields are masked before the AI model or the connector sees them, and each redaction is counted in the Security Posture reports.
05
What does the FinOps Guard actually do?
It truncates array responses after a configurable number of items, optionally compresses tool descriptions, and attributes exact token cost per execution against your cost per million tokens rate, turning AI Spend into a precise, per Connector ledger.
06
What happens when the Circuit Breaker trips?
When request rates exceed your budget ceiling inside the detection window, agent traffic is halted automatically and your budget is protected. The circuit auto resets after the cooldown period, or you can approve resumption manually at any time.
07
Is there an emergency stop for all agents?
Yes. The Global Emergency Halt deactivates every active Connector, revokes all tokens, and terminates all running sessions across your account. It requires typing HALT ALL to confirm, and connectors can be restored afterwards. Revoked tokens must be issued again.