Changelog

Built in the open.

Sabia shows the work your team's AI tools produced, the usage behind it, and whether the work was any good. Sabia is in early access. Here's every step, dated.

2026-10-02

Work Overview improvements

Work Overview is redesigned to lead with the artifacts, recorded outputs, and active projects. The output chart follows.
Work and Quality export as CSV for the period on screen, including a custom date range.
Outputs show who contributed, which tool, and which project.
Organizations can be renamed, left, or handed to a new owner. Empty pages say what will appear and how to get it.

2026-09-30

Claude Cowork, and deeper activity insight

Claude Cowork connects from Claude Desktop, including on Pro and Max plans, with usage credited to each person.
Activity Explorer replaces the tool-calls list. Trace any output or artifact back to the AI steps behind it, and see what each step took in and returned, by person, session, or tool.
Raw captures are deleted after a window each organization sets, from 7 to 90 days, 30 by default. Only owners and administrators can read them.

2026-09-24

Cost per project, and which work passes

A project page shows modelled AI spend, AI-assisted time, and the people involved for a period, plus a CSV export.
Quality Overview shows which checks pass, period by period, with the same kind of breakdown as Spend.
ChatGPT Work can report the work a session finished, through Output Reporting.

2026-09-17

Where the time in a session went

A session can show how long the model took, how long tools ran, and how long a person spent on an approval.
/why asks how agentic tools change the work, and follows Artifacts as they evolve.

2026-09-10

Traces behind the work

Sabia integrated Langfuse for trace observability, so Claude Code tool calls behind a piece of work are captured and can be traced.
The output dashboard is now Work Overview, and it opens on the day.
Billed and modelled spend stay on their own ledger, separate from that activity.

2026-09-03

A dashboard for the work

An output dashboard shows the most active project, artifact, and member for a period, with the output chart underneath.
Projects group that work. An Artifacts page lists what was produced.

2026-08-27

Checks against what happened next

Quality checks can draw on GitHub pull request reviews, merge state, and checks, and on Linear issue facts, so a check can reflect what happened to the work, not only that it was created.
The Output page can group that work by the artifact it belongs to.

2026-08-20

Outputs, and early access opens

An Output page lists work identified from Cursor, Codex, and Claude Code, by day, week, or month, and says so when nothing was identified.
An Evaluation page runs checks on those collected outputs. Cursor desktop usage includes token totals, not only a request count.
sabiapartners.ca goes live with the early-access list. Docs, changelog, blog, legal, and security pages are published, and contact is an on-site form.
Six connectors are committed for early access: OpenAI, Anthropic, and Cursor by provider API; Claude Code, Codex, and Cursor desktop by local telemetry. A number Sabia cannot see is reported as not reported, never as a fabricated zero.

2026-08-13

Claude Code, Cursor, and Anthropic

Connect Claude Code with the Sabia plugin. Usage from new sessions is recorded in Sabia.
Connect Cursor on the desktop. Sabia records each agent turn as activity: a request count, not token totals.
Anthropic usage and cost sync into Costs, alongside OpenAI. The report can be broken down and exported as CSV.

2026-08-06

The app opens

The Sabia app opens. Sign in, and Costs and settings sit in the sidebar.
Connect an OpenAI admin key and choose which projects Sabia reads. Usage and cost from that account sync in. An Anthropic account can be connected as well.
The Costs page shows spend and token usage for a period you choose, and how both moved over time.
Connect Codex on the desktop. Sabia records token use from those sessions, separate from billed API cost.