LatentWorkUsing LatentWork
Models, teams and usage
LatentWork uses the models your organization allows through your LatentStack gateway. Pick one per task, switch team or organization, and keep an eye on context and cost.
Where the model list comes from
Once LatentWork is connected to your gateway, the model list is exactly the models your tier allows, for the organization and team you've selected. To get a model added, ask your organization's administrator; see Admin Console.
Choose a model for a task
The model button under the message box shows the task's model. Select it, type in Search models... to filter, and pick one. The choice applies to this task only and is remembered for it.
New tasks start with your team's default model, as set by your administrator, or the first model on the list if there's no default. If you switch to a team whose tier doesn't include a task's model, the task moves to the new default.
Some models have variants, such as how much they reason before answering. For those, a Behavior menu appears next to the model. The model, organization and team can't be changed while the agent is working.
Selected model is unavailable. Choose another model before sending. Pick another model and send again.Organizations and teams
Your organization and team decide which models you can use, which limits and budgets apply, and where your usage is counted.
Team
The Teams button under the message box lists the teams you belong to, each with its tier. Choose one, or No team. The model list reloads for that team; a spinner shows while it does. The button then shows the team's name.
Organization
If your API key belongs to more than one organization, you can switch in three places:
- The organization button under the message box.
- The profile menu at the bottom left of the sidebar, under Organization.
- Settings → LatentRouter, under Organization. Default follows your default organization.
Switching organization clears the selected team and reloads the model list.
Changing team or organization takes effect right away. Changing the gateway URL or API key stops running tasks.
Context and cost
Once the first reply arrives, a ring appears at the right end of the row under the message box. It fills as the task uses up the model's context window, and turns red at 80%. Click it for the detail:
- How full the context is (… of context), in tokens and as a percentage, and the number of turns.
- The cost of the task so far, as reported by the provider, or cost not reported.
- A line per model used, with tokens split into in, out, cache r (cache reads), cache w (cache writes) and, when the model reports it, reasoning.
Cost comes from the provider, not an estimate. If the model's context window is unknown, no percentage is shown.
Running out of context
With Auto context compaction on (the default, in Settings → Preferences), the agent summarizes the earlier conversation by itself when the context fills up, and the task carries on. The summary keeps what matters; nothing is deleted.
With it off, LatentWork warns you at 80%, 90% and 95% with Context N% full. Type /compact and send it to summarize the task yourself.
Usage over time
Settings → Usage shows your spend through the gateway: Cost, Calls, Input tokens and Output tokens, then a table by model with cost, calls and tokens. Choose 7d, 30d (the default) or All.
| What you see | Meaning |
|---|---|
| Loading usage… | Fetching from the gateway. |
| No usage recorded yet. | No calls in that period. |
| An error message | The gateway couldn't be reached or refused the key. |
These are your account's figures from the gateway. For more detail, use the My Usage page in the console.