Integrations / Higress AI Gateway (direct)
Higress AI Gateway (direct)
Self-hosted Higress AI Gateway spend tracking via customer-side instrumentation. Higress (github.com/alibaba/higress) is an open-source (Apache-2.0), Envoy/Istio-based cloud-native API gateway that originated at Alibaba and is now a CNCF Sandbox project; deployed by the customer via Docker or Helm, never embedded by Telara. Its AI-proxy plugin does OpenAI protocol conversion in front of 100+ model providers, so a customer's OpenAI-compatible client pointed at their own Higress base_url (there is no single shared URL — customers supply their own) is instrumented the same way as any OpenAI-compatible endpoint. No first-party Higress wrapper exists in telara-observability; use the generic OpenAI wrapper against the customer's Higress base_url. pip extra: pip install telara-observability[openai] Because Higress is self-hosted with no Telara-visible vendor invoice, there is no bulk usage/billing API to pull at all — OTLP push from customer instrumentation is the only signal Telara can see. Higress sits between the customer and whichever underlying model vendor it proxies to, so cost attribution lands on the gateway hop as instrumented, not on the origin model provider — a real chargeback limitation, and here also an availability limitation: if the customer does not instrument, there is nothing to attribute.
What Telara does
Capability matrix
Before you connect
Prerequisites
Vendor setup
Get the credential
In Telara
Connect in Telara
This connector does not take a vendor key in Settings → Integrations. Run `telara otlp` (or point the vendor’s OpenTelemetry exporter at the endpoint that command prints) so usage arrives as telemetry.
Telara has not published the exact form fields for this method yet. Complete whatever the Connect dialog shows.
Spend
What gets measured
Resolution: Per request.
Attribution: Matched to people by email.
How far back vendor history goes is not published uniformly for this connector.
- input_tokenstokens
- output_tokenstokens
- framework_runscount
Sync
Freshness & sync
Spend connectors refresh on the vendor pull cadence declared in the catalog. Indexing lag is not published for this connector.
Data handling
Permissions & data handling
Telara has not published a permission list for this connector yet. Treat the vendor consent screen as the source of truth until this page lists scopes.
This catalog entry does not declare extra semantic-search exclusions.
After connect
Verify it worked
This connector does not declare a credential probe yet. A saved connection means Telara stored the credential, not that the vendor accepted it.
Honesty

