Portkey: 45 minutes to 2, across 200+ tickets.
Portkey is an AI gateway platform handling billions of API requests. Their support team was spending 20–45 minutes per ticket manually pulling data from ClickHouse, Linear, Stripe, and GitHub. After deploying Altor, median investigation time dropped to 2 minutes — with zero changes to their existing support workflow.
median diagnosis time (down from 20–45 min manually)
tickets diagnosed since deployment
production systems connected (ClickHouse, Linear, Stripe, GitHub, docs, StatusPage)
from kickoff to first live investigation
The problem
Portkey routes billions of LLM API requests through their gateway. When a customer reports "my API calls are failing" or "latency spiked since this morning," Portkey's support team needs to investigate across multiple systems: API logs in ClickHouse, known bugs in Linear, customer billing in Stripe, and recent deploys on GitHub.
Each investigation followed the same manual pattern: open ClickHouse, write a query, find the error pattern, switch to Linear, search for matching bugs, check Stripe, check GitHub deploys, cross-reference everything, and write the response. This took 20–45 minutes per ticket — for a workflow that was 80% identical every time.
Portkey wasn't lacking a support platform. They had one. What they lacked was the investigation layer — something that could pull live data from their actual systems and produce a diagnosis, not just route the ticket to the right person.
The deployment
Altor deployed at Portkey in a forward-deployed engagement over 3 weeks:
- Week 1: Stack audit. Mapped Portkey's systems, ticket types, and investigation workflows. Identified 3 primary ticket patterns (API errors, latency spikes, webhook failures) that accounted for 70%+ of support volume.
- Week 2: Connections live. Read-only integrations to ClickHouse, Linear, Stripe, GitHub, internal docs, and StatusPage. First investigations running on real tickets from the queue.
- Week 3: Playbooks tuned. Investigation logic refined against Portkey's actual ticket patterns. Support team trained on reviewing and sending Altor-generated diagnoses.
A real investigation: latency spike diagnosis
A Portkey customer submits: "Dashboard is loading slowly since this morning." Here's what Altor does in under 2 minutes:
- Queries ClickHouse: p95 latency jumped from 340ms to 1.2s starting at 09:14 UTC.
- Checks GitHub: Deploy #447 shipped at 09:12 UTC — introduced a new query in the /analytics endpoint.
- Searches Linear: No related bugs filed yet.
- Checks StatusPage: No upstream incidents. All providers green.
- Delivers diagnosis: Deploy #447 introduced an unindexed query in /analytics. Rollback or index addition will resolve.
"Altor diagnosed in 2 minutes what used to take our engineers 45 minutes of copying data between tabs."
The result
After 200+ tickets diagnosed through Altor, the numbers speak for themselves:
- •Median investigation time: 2 minutes (down from 20–45 minutes)
- •6 systems queried per investigation automatically — no more copying data between tabs
- •80% of investigation logic reusable across different ticket types
- •Zero changes to Portkey's existing support platform or team workflows
- •Support engineers shifted from manual investigation to reviewing and sending AI-generated diagnoses
Your stack looks like Portkey's. See what Altor finds.
We'll connect to your systems and investigate a real ticket from your queue — live. Same 2-minute diagnosis, on your data.
Get a 3-minute walkthrough — no call needed.