An investigation agent that lives in your IDE

Describe what's going wrong and Osuite's investigation agent finds the root cause and a fix — correlating logs, traces, and metrics across every service, right where you already work. Plus a visualizer agent that builds any dashboard from plain English.

your agent / checkout → inventory
Demo · sample data
Skillsosuite-investigate

User: Customers see a checkout timeout after pressing Pay. Is checkout the cause?

I’ll trace the failed request across services, then follow the slowest dependency.

Find slow checkoutsComplete
osuite traces search --service checkout-service --operation "POST /checkout" --min-duration 4000 --range 1h --limit 5
Follow the requestComplete
osuite traces getbyid a19c00f537cd4b098596a6801a99ba3c
Find the bottleneck evidence
SpanFinding
Browser → POST /checkout504 after 5,220 ms
checkout-service → inventory.reserve5,108 ms waiting downstream
inventory-service → UPDATE stock5,040 ms waiting on a row lock

Checkout is waiting on inventory. I’ll check the lock warning and the stock sync running at the same time.

Inspect inventory warningComplete
osuite logs search --filter resource.service.name__is=inventory-service --filter traceId__is=a19c00f537cd4b098596a6801a99ba3c --range 1h --limit 5
Trace the concurrent syncComplete
osuite traces search --service inventory-worker --operation sync_stock --range 1h --limit 5
Find the bottleneck evidence
EvidenceFinding
inventory-service logStock row lock wait exceeded 5,000 ms
inventory-worker · sync_stockTransaction open 62 s during vendor API call
Same SKU · same intervalStock sync holds the row needed by reserve
Proposed fix

The checkout 504 is a downstream symptom.

inventory-worker holds the stock row lock while it waits for a vendor API response. That blocks inventory.reserve until checkout times out. Fetch vendor data before opening the database transaction; review and test this change before applying it.

inventory_worker.py · proposed change
- with db.transaction():
-     db.select_stock_for_update(sku)
-     stock = vendor.fetch_stock(sku)
-     db.update_stock(sku, stock)
+ stock = vendor.fetch_stock(sku)
+ with db.transaction():
+     db.select_stock_for_update(sku)
+     db.update_stock(sku, stock)

Find the bottleneck: result ready.

Less time digging. More time fixing.

Investigation Agent

Lives in your IDE. Describe an issue and it correlates logs, traces, and metrics to surface the root cause — and a suggested fix — in plain language

Visualizer Agent

Build any dashboard by describing what you want, no PromQL or metric expertise required

Built into the platform

Not a bolt-on. Both agents work with all your existing Osuite data from day one

Works where you work

Your editor, your agent, your production context

One install puts the Osuite CLI and four agent skills where your coding agent already looks for them. Ask in plain language or start a skill as a slash command, in the same window where you write the fix.

  • Claude Code
  • Cursor
  • Antigravity
  • Any IDE with agent skills
curl -fsSL https://osuite.io/install-osuite-cli.sh | bash
CLI
~/.local/bin/osuite
Skills
~/.agents/skills · ~/.claude/skills
  • /osuite-investigate
  • /osuite-instrument
  • /osuite-edit-dashboard
  • /osuite-write-promql

Investigation Agent — Root Cause and a Fix, in Your IDE

  • Describe the issue in your IDE — the agent correlates logs, traces, and metrics across every service to find the root cause
  • Surfaces what broke, why, and which service is the origin — in plain language, with a suggested fix
  • Cuts investigation time from hours to minutes, even during complex cascading failures

User: The “Gateway error rate > 3%” alert is firing. Find the cause and suggest a fix.

I’ll find the service behind gateway’s errors, follow one failed request to its span and log, then check the code.

Find failing gateway requestsComplete
osuite traces search --service gateway --min-duration 5000 --range 1h --limit 5
Follow the slowest failureComplete
osuite traces getbyid 9c7ed92eb31b7689727afc425443b349
Read the orders error logsComplete
osuite logs search --filter resource.service.name__is=orders --filter severity.text__is=error --range 1h --limit 5
Investigate an issue evidence
EvidenceFinding
alert · Gateway error rate > 3%Firing 12m · 5.53 against a threshold of 3
GET /api/orders/:orderId · trace 9c7ed92e504 after 8.35s; SELECT orders failed after 4.32s
orders · error logpg-pool: timeout exceeded when trying to connect

User: Why is the pool running dry?

Inspect the query pathComplete
Read orders/orders.repository.js → getOrder()
Proposed fix

Root cause found. Fix ready for review.

getOrder() releases its pooled Postgres client only when the query succeeds. Every failed query leaks a client until the pool is empty, so orders times out waiting to connect and gateway returns 504s. Release the client in finally, so failed queries return it to the pool too.

orders/orders.repository.js · proposed change
  async getOrder(id) {
    const client = await this.pool.connect();
-   const { rows } = await client.query(ORDER_BY_ID, [id]);
-   client.release();
-   return rows[0];
+   try {
+     const { rows } = await client.query(ORDER_BY_ID, [id]);
+     return rows[0];
+   } finally {
+     client.release();
+   }
  }
Fix ready for review

Investigation Agent — Smarter On-Call

  • Does the first layer of investigation automatically, so on-call engineers arrive with context, not a blank screen
  • Delivers a full investigation summary — and a proposed fix — before a human even opens the incident
  • Works across your entire stack — no need to configure it per service or signal type
Osuite alert detail for the Gateway error rate > 3% rule: Firing for 12m at Critical severity, its threshold chart crossing 3, links into gateway in APM, error logs and failed traces, and a last evaluation of 5.53
The alert that pages on-call
  1. The alert that pages on-call

Visualizer Agent — Natural Language Dashboards

  • Describe any dashboard in plain English and Osuite builds it — no query language required
  • Automatically selects the right metrics, visualizations, and time ranges for your request
  • Any engineer on your team can build production-ready dashboards, regardless of experience level
Osuite dashboard assistant on the Service overview dashboard: asked to add an error rate panel, it checks its query and proposes an Error rate by endpoint panel, previewed on the canvas next to Apply to draft and Discard
The panel it proposes, before you apply
  1. The panel it proposes, before you apply

Visualizer Agent — Faster Iteration

  • Scaffold dashboards in seconds and refine them with the manual builder for full control
  • No more back-and-forth between engineers to find the right PromQL expression
  • Democratizes observability across your team — junior engineers are no longer blocked
The Service overview dashboard after applying: the new Error rate by endpoint panel sits under the services table, with endpoint error rates jumping late in the hour, ready to edit like any other panel
Applied, and yours to keep editing
  1. Applied, and yours to keep editing

Frequently Asked Questions

How does the investigation agent know where to look during an incident?
The investigation agent has access to all your telemetry signals — logs, traces, and metrics — across every service in Osuite. From your IDE, describe the issue and it correlates signals to build a timeline, identify the root cause, and suggest a fix — the same way a senior engineer would, just faster.
Does the investigation agent replace on-call engineers?
No — it makes them dramatically more effective. The investigation agent handles the first layer of investigation so your engineers arrive at an incident with context and a proposed fix, not a blank screen.
What if the Visualizer Agent picks the wrong metrics for my dashboard?
You can refine the result manually using the Custom Dashboard builder. Think of the Visualizer Agent as a fast starting point — you always have full control over the final output.
Do the AI agents require any special setup or configuration?
Only a one-time setup. One install adds the Osuite CLI and its four agent skills to your machine, and osuite login connects the CLI to your Osuite account through your browser. After that, the agents work with your existing Osuite data, with no model training and no per-service configuration.
Is my telemetry data used to train the AI models?
No. Your data is used only to power investigations and dashboard generation within your account. It is never used for model training.

See what Osuite can do for your team

Get a personalized walkthrough of the platform and a cost comparison against your current observability stack.

No commitment. No sales pressure. Just a real look at the platform.