FlowWatch — AI operations, under control

—/ 100

GLOBAL HEALTH

Waiting for telemetry

Connect a workflow to establish your operational baseline.

Runs + sessions · 24h—— success

Open incidents—Needs attention

AI cost · 24h—Workflows + agents

Risk exposure—Estimated open loss

7-DAY SIGNAL

Execution activity

Runs + sessions Failures

ACTION QUEUE

Recent incidents

BUSINESS CONTROL

What needs a business decision

Impact, spend forecast and the safest next action — based on real outcomes.

SPEND FORECAST

Budget runway

NEXT SAFE ACTION

Protect the outcome

EXTERNAL CREDIT RUNWAY

API balances and provider risk

DECISION LAYER

Command center

Group operations, inspect risk and move directly from signal to action.

Group by

GROUPED OPERATIONS

Risk by platform

live model

NEXT BEST ACTION

What needs a decision

UNIFIED ACTIVITY

Everything FlowWatch observed and changed

30-DAY RELIABILITY

Service level objectives

Within objective

AVAILABILITY—Waiting for operations

ALERT DELIVERY—Waiting for deliveries

MTTD · P95—Waiting for incidents

MTTR · P95—Waiting for resolutions

QUEUE LAG · P95—No active backlog

WORKFLOW PULSE

What is running now

Operational status, expected thresholds and the latest signal for every monitored automation.

Workflow	Platform	Health	Owner	Value / run	Last signal

Autonomous runtimes, their live health, cost envelope and recent decision sessions.

Agent	Runtime	Health	Channel	Last signal

LIVE TRACE

Recent agent sessions

Telegram and messaging bots observed from inbound update through agent reasoning to final delivery.

Bot	Runtime	Health	Channel	Last update

CONVERSATIONS

Recent bot sessions

Investigate, assign and resolve operational risk with a complete decision trail.

Raw execution telemetry with business outcome, cost and model context.

Workflow	Execution	Result	Duration	Cost	Business	Received

CONNECT YOUR RUNTIME

First signal in five minutes.

Use one scoped ingestion key for workflow events or the universal Agent Telemetry Protocol.

1
Create an ingestion key
The secret is displayed once and stored only as a secure hash.
2
Install a connector
Choose OpenClaw, Hermes, Telegram SDK or the n8n event contract.
3
Confirm business outcomes
Emit business.failed when a technically complete agent did not achieve its goal.

SECURE ACCESS

Ingestion API keys

ESCALATION

Notification channels

PROVIDER RUNWAY

External API balances

DELIVERY HEALTH

Alert outbox

RELIABILITY POLICY

Service objectives

ACCOUNT SECURITY

Sign-in protection

ACCESS CONTROL

Workspace team

REQUEST EXAMPLE

Send an event

Create an API key to generate your request example.

EVIDENCE, OUTCOMES & SPEND

Understand what changed — and what it cost.

Compare a run with healthy peers, explain health movement, connect changes to incidents and manage business outcomes.

HEALTH EXPLANATION

Why the score moved

—

TRACE COMPARISON

Compare with a healthy baseline

Select any recent execution or agent session. FlowWatch compares duration, cost and runtime behaviour without exporting sensitive content.

COST INTELLIGENCE

Spend, efficiency and guardrails

OUTCOME CONTRACTS

What success actually means

ANY BUSINESS DATA

Custom metric chart

Enter a metric sent by CRM, Stripe, Sheets or any webhook.

Choose a metric to beginUse names such as revenue, leads_created or refunds.

FLOWWATCH ADVISOR

Where spend and risk concentrate

Recommendations are based on your own tokens, cost, failures, outcomes and risky tool events.

TOKEN HOTSPOTS

MODEL HOTSPOTS

VULNERABLE RUNTIMES

CHANGE CORRELATION

What changed around the signal

DEPENDENCY MAP

See what each runtime depends on.

Model workflows, agents and notification channels so an incident includes its operational blast radius.

RESOURCES

Operational graph

—

AGENCY PORTFOLIO

Operate multiple client workspaces without mixing their data.

Portfolios aggregate only organizations where you already have administrator access.

QUALITY CONTROL

Version the decisions that make your agents useful.

Registry versions are safe metadata only. Evaluations connect quality evidence to real runs and sessions.

RELIABILITY COPILOT

What needs attention now

Evidence-backed

PROMPT & MODEL REGISTRY

Production versions

EVALUATION HUB

Quality checks

HUMAN CONTROL LAYER

Autonomy where it is safe. Approval where it matters.

Policies define when agents must pause. The inbox records a reviewable, auditable decision without executing arbitrary remote actions.

POLICY INBOX

Awaiting a human decision

APPROVAL POLICIES

Where agents pause

PLAN & BILLING

Give the operation room to grow.

Choose the capacity your team needs. A selected plan is recorded for this workspace; production limits change only after verified payment setup.

Loading current plan…

CURRENT CAPACITY

What this workspace is using

—

No charge is made from this screen while a payment provider is not configured. The plan choice is saved for the workspace owner and recorded in the audit log.

PRE-PRODUCTION SAFETY

Catch regressions before they reach production.

Synthetic checks observe a safe recent signal. Replay scenarios evaluate bounded historical telemetry without invoking your live workflow or agent.

SYNTHETIC MONITORS

Safe release sentinels

REPLAY SANDBOX

Historical contract checks

AGENCY OPERATING SYSTEM

Run every client workspace from one operating model.

Packs copy reviewed guardrails only to workspaces you administer. Every application is auditable.

CLIENT WORKSPACES

Where attention is needed

REUSABLE PACKS

Apply consistent guardrails

MCP GOVERNANCE

Know which tools an agent may use.

Register MCP servers, make tool-level decisions and identify missing ownership, authentication and policy coverage.

MCP SERVERS

Registered infrastructure

TOOL POLICIES