GitHub Copilot for JetBrains adds Ollama local models and memory
Run open weights in your IDE, keep context between chats, read your token bill line by line, and move off MAI-Code-1-Flash before Sept 10.
Copy markdown
Ollama BYOK lands in JetBrains
Copilot for JetBrains now accepts Ollama as a bring-your-own-key provider, so you can select a local model in the picker and keep code on your own machine. It is live in the latest plugin build — a real path to running open weights inside the IDE without a cloud round-trip.
Memory that survives the session
Copilot memory now retains and recalls project details across agent chat sessions, so you stop re-pasting your stack and conventions every conversation. Toggle it on or off in Copilot settings.
See every token you're billed for
The usage report now itemizes input, output, cache-read and cache-write tokens per model next to the AI credits each one burned — for Business and Enterprise admins and every Individual user. Download it from the AI usage page to trace exactly where your credits go.
MAI-Code-1-Flash retires Sept 10
The original MAI-Code-1-Flash shuts down September 10; move workflows and integrations to the newly-GA MAI-Code-1.1-Flash first. Enterprise admins must enable the replacement in policy settings before the cutoff.
Elsewhere: Vercel puts a paper trail on tokens
Vercel's Connect Observability shipped the same day with line-level visibility into the token lifecycle, plus filtering and log-drain forwarding — another sign tooling is moving from 'trust the credit meter' to per-token accounting.