The AI Usage dataset holds daily measurements of AI coding assistant usage: who is using AI tools, how much, on what, and to what effect. It brings every AI coding assistant onto a single schema so they can be reported side by side.

Only GitHub Copilot is connected today. The Anthropic Claude, Cursor, OpenAI Platform and OpenAI Codex integrations are coming soon. Fields those providers populate exist in the schema but are empty for now, and each field article says so.

Before building your first widget, read these two articles. AI usage records are shaped differently from every other Keypup dataset, and a report that ignores this will silently double count.

Definition: AI Usage Record Structure - what a record is, and the two fields that must appear in almost every filter.
Definition: AI Provider Coverage & Limitations - which providers are available, what each one does and does not publish, and the known limitations that look like bugs.

In short: every record covers one UTC day, and two fields say what kind of measurement it is. aggregation_level says whether the record counts a user, a repository or the whole organization. breakdown says which cut of the provider's report it came from. Always filter on one value of each - the cuts overlap, and summing across them counts the same activity several times.

Click on any field to access its documentation.

Identity & period

Field label	Field ID in dataset
ID (system)	`_system_id`
ID (source)	`id`
Created at	`created_at`
Updated at	`updated_at`

Measurement type

Field label	Field ID in dataset
Aggregation level	`aggregation_level`
Breakdown	`breakdown`

Actor

Field label	Field ID in dataset
Actor	`actor_username`
Actor type	`actor_type`

Surface & language

Field label	Field ID in dataset
Surface Ref	`surface_ref`
Surface Name	`surface_name`
Language	`language`

Model

Field label	Field ID in dataset
Model Ref	`model_ref`
Model Name	`model_name`
Model vendor	`model_vendor`
Model speed	`model_speed`
Model context tier	`model_context_tier`
Model auto-routed	`model_is_auto_routed`

Client

Field label	Field ID in dataset
Client	`client_name`
Client ID	`client_id`
Client version	`client_version`
Client plugin	`client_plugin`
Client plugin version	`client_plugin_version`

Repository

Field label	Field ID in dataset
Repository	`repository_full_name`
Repository name	`repository_name`
Repository owner	`repository_owner`
Repository visibility	`repository_visibility`

Measures - interaction

Field label	Field ID in dataset
Interactions	`interactions`
Requests	`requests`
Sessions	`sessions`

Measures - tokens

Field label	Field ID in dataset
Input tokens (uncached)	`input_tokens_uncached`
Input tokens (cache read)	`input_tokens_cache_read`
Input tokens (cache write)	`input_tokens_cache_write`
Output tokens	`output_tokens`

Measures - code outcome

Field label	Field ID in dataset
Suggestions offered	`suggestions_offered`
Suggestions accepted	`suggestions_accepted`
Suggestions rejected	`suggestions_rejected`
Lines suggested (added)	`lines_suggested_added`
Lines suggested (deleted)	`lines_suggested_deleted`
Lines added	`lines_added`
Lines deleted	`lines_deleted`
Commits	`commits`
Pull requests created	`pull_requests_created`
Pull requests reviewed	`pull_requests_reviewed`
Pull requests merged	`pull_requests_merged`
Review suggestions offered	`review_suggestions_offered`
Review suggestions accepted	`review_suggestions_accepted`

Measures - consumption

Field label	Field ID in dataset
Web search requests	`web_search_requests`
Credits used	`credits_used`

AI Usage Dataset Overview