Centralized Llm Model Configuration And Purpose-Scoped Allowlisting For Enrichment Workflows

Issue 91 Edition 2026-04-01 3 min read

Not accepted General

Sources: 1 • Confidence: Medium • Updated: 2026-04-12 09:58

Key takeaways

datasette-enrichments-llm has a new published release available as version 0.2a0.
datasette-enrichments-llm now uses datasette-llm to configure and manage LLM models.
Model availability for enrichments can be controlled by configuring models with an "enrichments" purpose.

datasette-enrichments-llm has a new published release available as version 0.2a0.
datasette-enrichments-llm now uses datasette-llm to configure and manage LLM models.
Model availability for enrichments can be controlled by configuring models with an "enrichments" purpose.

What specific behavior changes (including any breaking changes) were introduced in version 0.2a0 compared to the prior release?
What is the exact configuration interface in datasette-llm that the plugin now relies on, and how does it map to enrichment execution (e.g., selection UI/CLI, defaults, per-enrichment overrides)?
How is the "enrichments" purpose enforced (hard block vs. soft filtering), and what happens when no models are configured with that purpose?
Are there any additional purpose categories besides "enrichments" and any documented best practices for purpose design (e.g., cost tiers, data sensitivity tiers)?
Is there any direct decision-readthrough (operator, product, or investor) recommended by the source regarding adopting 0.2a0 now versus waiting for a stable release?

Shift to shared datasette-llm configuration suggests consolidation across the Datasette ecosystem, potentially reducing duplicate plugin work and increasing consistency for LLM features.
Purpose-scoped allowlisting for enrichments implies stronger governance controls for LLM usage, which could make enterprise or regulated adoption easier if it proves reliable.

Documentation and release notes show multiple plugins adopting datasette-llm as the common model configuration layer, with fewer plugin-specific settings over time.
User reports or examples demonstrate that purpose gating reliably limits enrichment model choices and provides clear behavior when no enrichments-purpose models are configured.

Version 0.2a0 introduces breaking or confusing behavior changes that increase operational burden versus prior releases, especially around model selection and defaults.
Purpose allowlisting is only a UI filter or can be bypassed, or failure modes when no enrichments-purpose models exist block workflows without clear guidance.