Every reply, regeneration, continuation, memory run and other AI request runs as a job. Jobs at /jobs shows the queue live so you can see what is waiting, what is stuck and what failed. Analytics at /analytics charts what those jobs cost and how fast your models are. Both are in the top bar.
Jobs
To find a job
- Open Jobs at
/jobs. Jobs are listed newest first and update live. - In the left pane, pick All or one chat. Each entry counts the jobs that still need something (queued, running or awaiting review). The Jobs button in a chat's header opens this page already narrowed to that chat.
- Narrow the list with the Status filter (Any status or one status), and with All jobs or Memory only to show every kind or memory jobs alone.
To stop jobs
- To stop one job, click Cancel on its row. Only queued and running jobs can be canceled.
- To stop everything in the current scope, click Stop all. You are asked Stop every queued and running job? first.
- To stop only memory work, click Stop memory jobs.
To retry a failed job
Click Retry on a memory job that failed, was canceled or was interrupted. For a scene memory, Retry memory only retries it without running the after-memory side prompts. Chat replies aren't retried here; regenerate them in the chat instead.
To review or dismiss a job
- A memory job set to wait for review shows Review, which opens the draft so you can save or discard it. See Memory.
- Dismiss removes a finished job from the list.
- Raw output expands the model's raw reply, which helps when a job produced nothing usable.
- Live output on a running job shows the text received so far, refreshed every second, so you can tell a slow job from a stuck one.
Each row's details name the model and, when there was one, the prompt that built the request.
Statuses
| Status | Meaning |
|---|---|
| Pending | Queued, waiting for a free slot on the connection. |
| Running | Sent to the model and streaming. |
| Awaiting review | Finished and waiting for you to accept the result. |
| Completed | Finished successfully. |
| Failed | Ended with an error, shown on the row. |
| Canceled | Stopped by you or by a script. |
| Interrupted | Cut off, for example by a server restart. |
Common failure messages include Stopped: the stream went silent., Stopped: no tokens arrived before the first-token timeout., Stopped: the reply exceeded the output cap., Stopped: the model kept repeating itself. and No enabled connection serves this model. Timeouts come from the model's job time limits; see Models.
Job kinds
| Kind | What it is |
|---|---|
| Reply, Regenerate, Continue, Refine | Chat generations. |
| Memory, Consolidation, Compaction, Side prompt, Topical clip, Memory assistance | Memory work. |
| Caption | Describing an image for models that can't see images. |
| Expression | Classifying a message's expression. |
| Assistant | A reply in a Chat with Mythweaver conversation. |
Note
Finished chat generations are deleted from the list after a day, and other finished jobs after 30 days. Analytics keeps its own records, so the charts are not affected.
Analytics
Analytics at /analytics charts your job runs per day in your browser's time zone. Every job run counts: replies, regenerations and memory work.
To change what the charts show
- Pick a range of 7 days, 30 days (the default) or 90 days.
- Switch between Mine and All users. All users is available to administrators only.
The page reloads when one of your jobs finishes.
Charts
| Chart | What it shows |
|---|---|
| Tokens per day | Prompt and completion tokens per day, by model. |
| Cached and uncached prompt tokens per day | How much of the prompt the backend served from its prompt cache. Cached prompt tokens are cheaper and faster on providers that support caching. |
| Tokens by model | Each model's total split into cached prompt, uncached prompt, prompt with cache not reported, and completion. |
| Spend per day | Cost per day by model. When the connection reports no cost, it is estimated from the model's prices; a note says how many runs had a known cost and how many were estimated. |
| Median time to first token | Per model, with a whisker to the 95th percentile. Timed from when the job started running, so time spent in the queue is left out. |
| Generation speed | Median completion tokens per second after the first token, per model. |
| Outcomes by job type | Runs by job kind and how they ended: Completed, Awaiting review, Failed, Canceled or Interrupted. |
A chart with nothing to show in the range shows a note instead, such as No completed runs in this range. Set model prices on the model's page to get spend figures; see Models.