Add a local GPU path that can execute image generation on an existing user GPU through a supported local backend/API. Prefer integration with an established server/workflow interface over embedding model runtimes or downloading weights from Archie.
Ship a tested initial recipe set for common models appropriate to consumer GPUs, with the exact model files, backend versions, VRAM/RAM expectations, supported operations, and known limitations recorded. Include runnable code/config examples that Archie can use to assist a user in setting up and calling the backend.
The implementation must report unavailable hardware/backend/model conditions clearly and must not pretend an untested model is supported.
Acceptance criteria
At least one local GPU recipe is exercised end-to-end on real hardware or a reproducible maintained test harness.
Additional recipes are included only with evidence, covering model/backend/version, VRAM, image sizes, and speed/quality trade-offs.
Backend URL/auth/timeout and model selection are configurable.
Local generation returns the same provider-neutral result shape as hosted generation.
No model weights are committed or downloaded into the repository/home partition.
Design
See docs/prds/image-capability-contract.md#local-gpu-provider-499 for the settled design: ComfyUI as the baseline recipe (submit/poll shaped like minimax.Client.GenerateAndWait), config shape, and capability mapping.
Suggested /council lenses
lens-operator, lens-deletionist -- run /council --lenses operator,deletionist once gate-clean, before opening the PR.
Close reason
Closed
Imported from beads archie-core-1786748942219-5-9844a72e · GitHub #499 · type task · priority P2 · created 2026-08-14 · closed 2026-10-05
## Scope
Add a local GPU path that can execute image generation on an existing user GPU through a supported local backend/API. Prefer integration with an established server/workflow interface over embedding model runtimes or downloading weights from Archie.
Ship a tested initial recipe set for common models appropriate to consumer GPUs, with the exact model files, backend versions, VRAM/RAM expectations, supported operations, and known limitations recorded. Include runnable code/config examples that Archie can use to assist a user in setting up and calling the backend.
The implementation must report unavailable hardware/backend/model conditions clearly and must not pretend an untested model is supported.
## Acceptance criteria
- At least one local GPU recipe is exercised end-to-end on real hardware or a reproducible maintained test harness.
- Additional recipes are included only with evidence, covering model/backend/version, VRAM, image sizes, and speed/quality trade-offs.
- Backend URL/auth/timeout and model selection are configurable.
- Local generation returns the same provider-neutral result shape as hosted generation.
- Tests cover backend unavailable, malformed response, timeout, cancellation, and model-not-found paths.
- No model weights are committed or downloaded into the repository/home partition.
## Design
See docs/prds/image-capability-contract.md#local-gpu-provider-499 for the settled design: ComfyUI as the baseline recipe (submit/poll shaped like minimax.Client.GenerateAndWait), config shape, and capability mapping.
## Suggested /council lenses
lens-operator, lens-deletionist -- run `/council --lenses operator,deletionist` once gate-clean, before opening the PR.
### Close reason
Closed
---
_Imported from beads `archie-core-1786748942219-5-9844a72e` · GitHub #499 · type task · priority P2 · created 2026-08-14 · closed 2026-10-05_
sam
added the mediumtask labels 2026-10-27 13:34:38 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Scope
Add a local GPU path that can execute image generation on an existing user GPU through a supported local backend/API. Prefer integration with an established server/workflow interface over embedding model runtimes or downloading weights from Archie.
Ship a tested initial recipe set for common models appropriate to consumer GPUs, with the exact model files, backend versions, VRAM/RAM expectations, supported operations, and known limitations recorded. Include runnable code/config examples that Archie can use to assist a user in setting up and calling the backend.
The implementation must report unavailable hardware/backend/model conditions clearly and must not pretend an untested model is supported.
Acceptance criteria
Design
See docs/prds/image-capability-contract.md#local-gpu-provider-499 for the settled design: ComfyUI as the baseline recipe (submit/poll shaped like minimax.Client.GenerateAndWait), config shape, and capability mapping.
Suggested /council lenses
lens-operator, lens-deletionist -- run
/council --lenses operator,deletionistonce gate-clean, before opening the PR.Close reason
Closed
Imported from beads
archie-core-1786748942219-5-9844a72e· GitHub #499 · type task · priority P2 · created 2026-08-14 · closed 2026-10-05