Ollama local AI models provider for the Backdrop CMS AI module.
Installation
- Install this module using the official Backdrop CMS instructions.
Model capabilities
The provider lists models through /v1/models, then reads each model's native
/api/show capabilities to filter selectors: completion for chat/text,
embedding for embeddings, vision for vision, and tools for tool calling.
Model names are not used to guess capabilities. The base URL may include /v1.
Capability lookups also recognize thinking, insert (fill-in-the-middle), and
audio metadata. Vision, tool calling, thinking, and insertion additionally
require completion. Audio metadata does not imply dedicated TTS/STT endpoint
support. Native metadata does not enable currently unimplemented image, speech,
or moderation methods in this adapter.
Models with the decision capability (Ollama 0.35 or later) are listed as
decision models, and decide() sends them to Ollama's /v1/systemone
endpoint for typed answers with calibrated probabilities. Other models answer
decisions through the AI module's tool-calling or JSON emulation.
Metadata is cached for six hours; failed or missing metadata is cached for five
minutes. The AI settings model-cache refresh clears this metadata too. Discovery
does not run inference. Servers or proxies must expose /api/show for automatic
classification. Models without metadata remain in the full catalog but are
excluded from capability-specific selectors. Existing manual overrides at
admin/config/ai/settings/capabilities/ollama and capability alter hooks still
apply. Unsupported image, speech, and moderation operations have no automatic
model selections.
Run offline regression checks with php tests/capabilities.php from this module.
Issues
Bugs and feature requests should be reported in the Issue Queue.
Current Maintainer
Credits
-
Created for Backdrop CMS by Justin Keiser.
-
Developed with AI assistance.
License
This project is GPL v2 software. See the LICENSE.txt file in this directory for complete text.