0ff0202128
* feat(llama-cpp): add managed local OCR and vision setup * test(llama-cpp): clarify synthetic endpoint credential fixtures * fix(llama-cpp): preserve media across managed setup * docs: refresh config baseline * test: isolate native output fidelity fixture * test(ui): await avatar loader completion * style(ui): format avatar loader test * chore: restore focused local media scope * refactor(llama-cpp): separate runtime inspection from preparation Keep managed server preparation within its size budget and preserve upstream diagnostic behavior. Retain real media SDK exports in tests and satisfy catalog lint checks. * fix(llama-cpp): enforce media budgets over inherited router limits --------- Co-authored-by: Peter Steinberger <steipete@gmail.com>
1.7 KiB
| summary | read_when | title | |
|---|---|---|---|
| Managed and external llama.cpp servers for GGUF chat, embeddings and local images. |
|
Llama Cpp plugin reference |
Managed and external llama.cpp servers for GGUF chat, embeddings and local images.
Distribution
- Package:
@openclaw/llama-cpp-provider - Install route: npm or ClawHub:
clawhub:@openclaw/llama-cpp-provider
Surface
- Providers:
llama-cpp - Contracts:
embeddingProviders,mediaUnderstandingProviders,tools
Managed text models
During interactive setup, OpenClaw installs a pinned, verified llama-server
and recommends a model for the Gateway host's available memory, GPU, and disk
space. Recommendations start with Qwen3.5 4B (about 2.7 GB) on eligible 8 GiB
hosts and scale up as hardware permits. See the current
model recommendations for sizes and
selection floors.
To use another model, configure params.modelPath, select its llama-cpp/<id>
reference, and rerun managed setup. Setup uses that authored route and offers
to download it if needed. Custom models are not subject to recommendation
memory floors. Existing cached GGUFs remain supported; local memory search can
also use embedding-only setup.