Register a new app
The key is shown once at creation -- store it in the app's env vars. Most apps should enrol themselves instead: use Copy prompt below and hand it to whatever coding agent built the app. This form is the manual fallback.
The Dispatch slug ties this app to its entry in Dispatch's AI-provider export, so Atlas can advisory-flag drift between what an app declares it uses and what it actually calls through the gateway. Leave blank if the app isn't tracked in Dispatch yet.
Registered apps
1 app · 0 tasks
| Name | Dispatch slug | Key prefix | Tasks | Status | Created | |
|---|---|---|---|---|---|---|
| atlas-smoke-test | — | atlas_live_rzI9I… | no slug — cannot match tasks | Active | 7/28/2026 |
Not yet on Atlas
29 apps Dispatch knows use an LLM and hold no Atlas key · 116 call sites the gateway could serve today, across the 24 that have been scanned. Copy an app's prompt and hand it to whatever coding agent built it — the code inside works only for that app, and expires in 7 days.
| App | Tasks | Call sites | Moves now | Stays, and why | |
|---|---|---|---|---|---|
| 18 | 29 | 20 | 6 meta, 2 image, 1 rerank | ||
| 16 | 30 | 14 | 9 image, 6 video, 1 self-hosted | ||
| 10 | 10 | 10 | — | ||
| 6 | 10 | 10 | — | ||
| 5 | 9 | 9 | — | ||
| 8 | 30 | 8 | 21 audio, 1 self-hosted | ||
| 14 | 21 | 8 | 8 audio, 2 embedding, 1 image, 1 meta, 1 self-hosted | ||
| 5 | 7 | 6 | 1 audio | ||
| 5 | 5 | 5 | — | ||
| 2 | 3 | 3 | — | ||
| 2 | 3 | 3 | — | ||
| 2 | 3 | 3 | — | ||
| 3 | 3 | 3 | — | ||
| 1 | 3 | 3 | — | ||
| 1 | 2 | 2 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
| 1 | 1 | 1 | — | ||
logisticsagentv2holds its own provider keys | — | — | not scanned | declares Anthropic — the agent will inventory it | |
sonanceapautomationholds its own provider keys | — | — | not scanned | declares Anthropic — the agent will inventory it | |
sonanceslackragholds its own provider keys | — | — | not scanned | declares Anthropic, OpenAI — the agent will inventory it | |
iportioagent2 | — | — | not scanned | declares Cortex — the agent will inventory it | |
roadmapvisualizer | — | — | not scanned | declares Cortex — the agent will inventory it |
47 more call sites sit in 4 repos that belong to no app profile in Dispatch —
peoplesoft_analysis (24), Elmo (16), iportio-agent (4), phase-gate (3). They cannot be prompted from here: a prompt is addressed to an app slug and these have none. Link the repo to an app in Dispatch and it will appear above on the next hourly sync.What Atlas can't route yet
61 of 177 scanned call sites stay where they are after every app migrates, plus 47 in repos linked to no app. Each row is one capability Atlas does not have — this is the queue behind the queue.
| Kind | Call sites | Apps | Reason an agent writes | Where |
|---|---|---|---|---|
| audio | 30 | 3 | non-text-output | unitree (21), athenav2 (8), sonanceorderautomation2 (1) |
| image | 12 | 3 | non-text-output | bloom (9), cortex (2), athenav2 (1) |
| meta | 7 | 2 | non-completion-endpoint | cortex (6), athenav2 (1) |
| video | 6 | 1 | non-text-output | bloom (6) |
| self-hosted | 3 | 3 | self-hosted | athenav2 (1), bloom (1), unitree (1) |
| embedding | 2 | 1 | non-completion-endpoint | athenav2 (2) |
| rerank | 1 | 1 | non-completion-endpoint | cortex (1) |
| unlinked repo | 47 | 4 repos | not a capability gap — link the repo in Dispatch | peoplesoft_analysis (24), Elmo (16), iportio-agent (4) +1 more |
Three things are missing from this table, and they are not small. Dispatch's scan records what a call produces, not how it is made — so streamed responses, tool calls and multimodal input are invisible here, and Atlas cannot route any of them either. Every one of them sits inside the 116 sites counted as movable above. The real figure is worse than this table shows, and the only way to learn it is to migrate an app and see what its agent marks blocked.