Documentation › Your instance
Your instance
Where your instance lives, what Thalamus runs for you, and what to check when something goes wrong.
Your instance’s address
Each ThalamusCUA instance on Thalamus AI Cloud has its own address:
https://<instance>.<environment>.thalamuscua.thalamusaicloud.com
For example, https://starcraft.staging.thalamuscua.thalamusaicloud.com is the instance starcraft in its staging environment.
The console and the API share that address:
| What | Where (for the example instance) |
|---|---|
| Console | https://starcraft.staging.thalamuscua.thalamusaicloud.com |
| API | https://starcraft.staging.thalamuscua.thalamusaicloud.com/v1 |
| API reference | https://starcraft.staging.thalamuscua.thalamusaicloud.com/docs |
What Thalamus runs for you
Everything behind the console runs inside your instance and is run by Thalamus — nothing to install, start, stop or upgrade:
- The API and the console.
- The workers that carry out goals, and the browser sandboxes the agents drive — one per agent run.
- The vision and embedding models and the servers that run them.
- The database and file storage that keep goals, plans, runs, traces and screenshots.
Your administrators manage what is yours to decide: accounts, environments and their logins, and which model fills each role.
Troubleshooting
| Symptom | Likely cause |
|---|---|
A write is refused, and env.not_disposable is logged | The environment is shared; use a disposable one. |
| A run sits BLOCKED | An approval is waiting on the Approvals page, or one was rejected. |
| The agent asks instead of acting | It has no verified skill for that operation, or a selector is missing; see Application briefs. |
| An evaluation is marked stopped early | The application instance stopped answering its health check, or the agent’s login no longer works there. Check the environment’s address and login. |
| The agent can’t open the application | Check the environment’s address, and that the application is reachable from your ThalamusCUA instance. |
| Planning times out | A thinking model is in the planning role; an admin can choose an -instruct one on the Configurator. |
| Steps take 10 s or more | The model in that role is slow on your instance’s hardware; the Configurator’s measured results show the step time of each model. |
| A run stays queued and never starts | Contact Thalamus support with the run’s id. |
The deeper evidence is always in the trace: the run’s report (report.md), or the
Trace explorer for every observation and screenshot. Include the run’s id
and its report when you contact Thalamus support.