Skip to main content
By the end of this guide, an app that Forge did not build and does not deploy sends its traces, logs and metrics to Reliant, and you can read them with reliant forge obs. It can be in any language, on any infrastructure: you add an OpenTelemetry SDK and a few environment variables. The variables are the standard OpenTelemetry ones, so there is no Reliant SDK to install. The token decides which organization and environment the data lands in.

Before you begin

  • A Reliant account in the organization the telemetry should belong to, with the obs:write permission to mint tokens. Owners and admins hold it.
  • The reliant CLI v1.21.3 or later, which includes forge v1.6.1 (install). Your repository does not need a forge.yaml.
  • Your own terminal and browser. Ingest tokens are minted by a person, so this cannot be run by an agent or a CI job.
1

Sign in as yourself

terminal
This opens your browser and stores a credential for Reliant’s control plane.
2

Register the environment and mint a token

From the root of your app’s repository, name the environments and the project to create or reuse:
terminal
In a repository with no Forge files, it asks before writing deploy/envs/prod.yaml (--write-config skips the question, --no-write writes nothing). Commit that file: every obs command reads the project from it.
deploy/envs/prod.yaml
It then prints the ingest token’s name, id and expiry date, followed by an export line for each variable. The token is shown once and expires after 90 days at most.
3

Set the variables where the app runs

Store the header as a secret in your deploy system, never in the repository, and set the rest beside it:
terminal
Use the values obs init printed. Give each deployable its own OTEL_SERVICE_NAME. Set service.version to what you deploy, such as the git SHA or release tag: the issue list’s “first seen in” and obs check both key on it. Keep deployment.environment.name equal to the env name you passed to obs init.
4

Add an OpenTelemetry SDK

Any SDK that reads the standard variables works. Zero-code instrumentation is the shortest path:For errors to become issues, failed requests must set the span status to error and record the exception. The auto-instrumentation above does this for unhandled errors. Export logs over OTLP (OTEL_LOGS_EXPORTER=otlp) only if nothing already ships the app’s stdout, or every line arrives twice.
5

Check that telemetry arrives

Deploy the app with the variables set, send it some requests, then:
terminal
Before anything arrives, the env exists but is empty:
Once data arrives, the telemetry line shows when the newest log and span came in, and your service appears with its request rate, errors and p95. Then read the data itself:
terminal
6

Rotate the token before it expires

Nothing renews a token. list marks one in its last 14 days with “expires soon: rotate it”:
terminal
rotate mints a replacement, and the old token keeps working for the grace period while you roll out the new header. obs tokens create prod --name <label> mints a separate token per cluster or CI system, and obs tokens revoke <token-id> --reason leaked cuts one off within minutes.

What this does not do

  • It does not build or deploy your app. The environment is registered as self-managed, and the platform never deploys there.
  • There is no anonymous path. Reading telemetry needs a sign-in, and minting tokens needs a person with obs:write. A CI job can read with a token that has obs:read (credentials).
  • With no deploy recorded, obs check judges a new service.version from the first time telemetry showed it, and marks the result marker INFERRED. A canary or slow rollout moves that moment.
  • A monitoring-only environment gets the built-in alerts (new and regressed issues, quota) and your organization’s alert receivers. Per-environment alert rules come from a Forge-deployed environment’s declaration (dashboards and alerts).
  • The ingest token is a server-side secret. Never ship it in a browser bundle or a mobile app.

Troubleshooting

obs init ran with an agent’s or a pipeline’s credential. Run it yourself in your own terminal after reliant forge cloud login. The refused run may still have registered the env and written deploy/envs/prod.yaml; running it again reuses both and only mints the token.
Your account lacks obs:write. Ask an owner or admin of the organization to grant it, or to run obs init for you.
Check that the app’s process actually has the variables, the protocol is http/protobuf, and the header reads authorization=Bearer <token> with no extra quotes. Check that the token is not revoked (obs tokens list prod --all).
That env is deployed by Forge, which wires its telemetry for you. For a token for an app outside Forge’s deploys, use reliant forge obs tokens create prod.

Next steps

Find which release introduced an error

Trace an issue to its release and suspect commits.

Bring an existing app to Forge

Go further: let Forge build and deploy the same app.

Debug production with an AI agent

Point an agent at the same logs, traces and issues.

Where the telemetry goes

Retention, usage and what the platform collects.