CHANGELOG

What shipped.

Features, improvements, and fixes — newest first, recorded as they ship.

Team workspaces, in beta

Team workspaces enter beta: share tasks, reports, and credits — one workspace for the whole research group

Release media (placeholder)

Hand any task off: every member can inspect the agent's full search trail, reading log, and edit history
Workspace-level billing, metered at official list prices, with monthly exports for admins
Open to all members during beta — create one from Settings in a click

Tasks running past six hours now checkpoint automatically and resume from the interruption, never from zero

Release media (placeholder)

Fixed deep search re-fetching the same sources, noticeably improving source diversity in long reports
The task page now streams live steps: what the agent is doing, which model it is using, and what it has spent
Fixed an occasional citation misalignment in PDF exports

Rebuilt the citation panel: every claim carries a clickable citation that jumps back to the exact source passage

Release media (placeholder)

Auto routing is live: every step of a task is routed across Claude Sonnet 5, Claude Opus 5, and GPT 5.6

Release media (placeholder)

Search and reading go to Claude, pure logical reasoning to GPT — which model ran each step is always visible
Routing decisions are written to the task log, replayable step by step
You can still pin a single model manually, metered at official list prices

Give Rarvis one research brief; a cloud agent searches, reads, analyzes, and drafts until it delivers a file you can cite

Release media (placeholder)

Launch deliverables: research reports, literature reviews, data tables, and decision briefs
Membership dues convert to credits at full value, metered at official list prices — we cover most of the bill
Free to start, no credit card, your first task runs in minutes

Only what has shipped makes this list.