One dashboard for AI cost and behavior across every coding tool your team uses and every AI feature in your product. Real-time anomaly alerts. Install in under an hour.
Session-volume anomaly · meeting-notes
5 Claude Code agents, 30s dispatch loop, running 10d unattended · 6,352 cycles · projected $400+
Claude Code and Cursor run on every engineer’s laptop. AI features run inside your product. No single dashboard sees both. You find out about runaway pipelines and spend spikes from the next invoice, days or weeks too late.
Questions every VP of Engineering, Head of Platform, and VP of Product needs to answer this week, and can’t:
Your AI runs in two places. AgentWatch covers both, with the same data model and the same view.
Every coding agent on every developer machine.
Every AI feature in your product.
One dashboard. Both surfaces. Same data model.
AgentWatch is not LLM observability. We don’t trace every model call or evaluate output quality. We answer cost, attribution, and behavioral anomaly questions across both your internal AI tools and your product’s AI. Most teams need both kinds of tools.
The shift, in numbers
of AI-native engineering teams now run 3+ AI models in production.
23% run 6+. Most have no per-model, per-team, or per-feature cost attribution today.
Source: Datadog, State of AI Engineering 2026
of platform engineering and SRE leaders will own production AI systems by 2028.
Most aren't equipped to manage the cost and behavior side yet.
Source: Gartner, Innovation Insight: LLM Observability, July 2025
Run one command on a developer machine. Copy your API key. Open the dashboard.
Install the collector
One command. Reads session files from Claude Code, Cursor, Codex, Cline, Windsurf. No code changes.
Optional: instrument product AI
One-line SDK per AI call. Python and JavaScript. Same data model as the collector.
Open the dashboard
Full AI cost and behavior, internal and product, in one view. Set alerts. Export to Slack.
API key shown once. Copy it now.
aw_live_dd0276952bb3c71b7225ab30a2a4457495113438e6d123741cf12cc0ba42a4e1Set as AGENTWATCH_API_KEY in your environment.
No infrastructure. No procurement. No instrumentation work. Run a command on a developer machine, open the dashboard, see your team’s AI usage.
A real user story
Five Claude Code agents on a 30-second loop. Spawned by a meeting-notes pipeline our team built. Ran for 10 days unattended. Generated 31,762 session files across 6,352 dispatch cycles.
AgentWatch fired a Slack alert the moment the anomaly crossed threshold. From the alert, we drilled into the session timeline, identified the runaway agent structure, and set a budget cap to prevent recurrence. Detection time: minutes. Without the collector, the bill would have shown up two weeks later as a $400 mystery.
Session-volume anomaly detected
Project meeting-notes is running 5 Claude Code agents on a 30-second loop. Unattended for 10 days. 6,352 dispatch cycles. Projected cost $412.
Agents
5
Sessions
31,762
Projected
$412
The alert that fired in our team’s #eng-alerts channel.
The detection layer
AgentWatch watches for a small set of behaviors that go wrong with agents, and gives you a clear move for each one.
A project's session count diverges from its 7-day baseline. The pattern that catches runaway pipelines.
Same prompt or tool-call signature repeats beyond a threshold within a session. Catches agents stuck on a sub-task.
Cost-per-hour ramping faster than the project's baseline. Catches expensive-model misuse before the invoice.
Long-running session with no human input for hours. Catches the 'left it running over the weekend' case.
What we did when the meeting-notes anomaly fired.
Slack alert fires in #eng-alerts: session-volume anomaly · meeting-notes.
Open the session timeline from the alert. Five agents on a 30-second loop.
Pause the project and set a $50/day budget cap from the alert itself.
Promote the diagnosis to a rule: any project crossing 100 sessions/hour auto-pauses.
Four roles. One dashboard.
VP Engineering & Platform
You own the AI tool seats. You don't own the answers your CFO and team are asking.
One dashboard for every coding agent on every developer machine.
See the VP viewVP Product
Your CFO asked what AI costs per customer. You don't have the dashboard.
Per-feature, per-user, per-customer cost attribution.
See the VP viewFinance & FinOps
AI tooling is now a real budget line. It moves like a startup.
Forecasted AI spend with weekly variance reports.
See the Finance viewEngineering Managers
Your team is using more AI tools than you can track.
See how each engineer uses every AI tool.
See the Engineering viewWe’re working with a small number of design partners in 2026. If your team is running multiple AI tools and feels the cost and visibility gap, we’d like to talk.
We reply within 48 hours. No newsletter, no marketing automation. Just a real reply from one of us.