Release Milestone And New Extension Surface

Issue 84 Edition 2026-03-25 5 min read

General

Sources: 1 • Confidence: High • Updated: 2026-04-13 03:53

Key takeaways

datasette-llm version 0.1a1 has been released.
Dependent plugins can request a model by purpose via an API (for example purpose="enrichment") and can register their purpose strings via register_llm_purposes() to enable centralized purpose listings for features such as an admin UI.
datasette-llm is a base plugin that makes models from the LLM project available to other Datasette plugins, including datasette-enrichments-llm.
datasette-llm 0.1a1 adds a register_llm_purposes() plugin hook and a get_purposes() function to retrieve registered purpose strings.
datasette-llm centrally configures which models are used for different purposes, such as data enrichment versus SQL query assistance.

datasette-llm version 0.1a1 has been released.
datasette-llm 0.1a1 adds a register_llm_purposes() plugin hook and a get_purposes() function to retrieve registered purpose strings.

Dependent plugins can request a model by purpose via an API (for example purpose="enrichment") and can register their purpose strings via register_llm_purposes() to enable centralized purpose listings for features such as an admin UI.
datasette-llm centrally configures which models are used for different purposes, such as data enrichment versus SQL query assistance.

Dependent plugins can request a model by purpose via an API (for example purpose="enrichment") and can register their purpose strings via register_llm_purposes() to enable centralized purpose listings for features such as an admin UI.
datasette-llm is a base plugin that makes models from the LLM project available to other Datasette plugins, including datasette-enrichments-llm.

What is the concrete configuration format and precedence model for mapping purposes to models (e.g., where it lives, how overrides work, and how defaults are chosen)?
What standardized purpose strings exist or are recommended, and how are naming collisions or conflicting semantics across plugins handled?
Which models from the LLM project are supported through datasette-llm in practice, and what are the operational constraints (latency, quotas, cost visibility) exposed to administrators?
Is there an implemented admin UI (or equivalent management surface) that consumes get_purposes(), and how widely is it used in real deployments?
What are the security and governance controls associated with purpose-based model selection (e.g., preventing unapproved purposes/models, logging/auditing by purpose)?

Purpose based model routing could lower integration friction across Datasette plugins, increasing adoption of an LLM enabled plugin ecosystem if configuration is clear and centrally managed.
A standardized purpose registry may enable admin tooling such as UIs and governance features, suggesting a pathway to enterprise readiness if the ecosystem converges on shared purpose strings.
Positioning datasette-llm as a base layer could concentrate model access and configuration, implying downstream plugins may align on this interface rather than bespoke LLM integrations.

Documentation of configuration format and precedence for mapping purposes to models, including defaults and override behavior across deployments.
Evidence of ecosystem adoption such as multiple plugins using purpose requests and registering purposes, plus a discoverable list via get_purposes in real deployments.
An implemented admin management surface consuming get_purposes, alongside stated controls for security, auditing, and cost or quota visibility by purpose.

No clear or stable configuration and precedence model emerges, causing downstream plugins to hard code models or implement parallel configuration paths.
Purpose strings fragment or collide across plugins with no recommended standards, reducing interoperability and limiting usefulness of centralized listings.
Operational constraints such as latency, quotas, or cost visibility are not exposed to administrators, limiting deployment beyond experimentation.