→ release notes · page 2

Changelog

Every trau release, newest first: what’s new, what changed and what got fixed.

Latest release

v8.2.0

Oct 15, 2026

Releases shipped

129

since June 2026

  1. v4.0.0

    Trau 4.0: Copilot, session history and cross-repo runs

    GitHub Copilot joins as a fifth provider, past agent sessions can be continued, and Workspace runs can request work from a sibling repo. Complexity bands now map to a Preset per provider.

    Added

    • GitHub Copilot CLI is a fifth provider for runs, interviews, the Assistant and the Terminal
    • Session history lists past agent conversations of every signed-in account, with details and a one-click continue in the Terminal
    • A run in a Workspace can file the ticket it needs in a sibling repo and picks up again once that ticket ships

    Changed

    • Models by complexity maps each provider to its own Preset per band; a band mapping set outside the hub must be chosen again on the Models page
    • Checklist checks tick proven steps automatically, and Trau can carry out a checklist step for you

    Fixed

    • Azure DevOps assigns filed work items to their owner again, and a split User Story files its slices as Tasks

  2. v3.0.0

    Trau 3.0: built-in prompts, benchmarks and a phone layout

    Prompt editing and Suggest as you type are gone, so every run uses the prompts trau ships. Interviews preview their results before approval, performance-sensitive tickets get a measured budget, and the hub fits a phone.

    Added

    • Performance-sensitive tickets get a performance budget in the Interview, the build measures it, verify checks it and the pull request reports the result
    • Interviews open with a restatement you accept, preview the exact result before approval, and let you change a result without losing reviewed work
    • Removing a project or repo lists what will be deleted, requires typing its name, and purges every record and file the hub kept for it

    Changed

    • The hub fits a phone: one page gutter, flat lists and controls first, with banners and desk-only detail hidden on small screens
    • Folder repos serve child apps on sibling URLs, warn about ignored folder-level app settings, and detect dev-server crash loops

    Removed

    • Prompt editing and Suggest as you type are gone; every run uses the prompts trau ships, and two settings turn the design rules and font policy off

  3. v2.73.0

    Cobble is the new look of the hub and the terminal UI

    The hub and the terminal UI take the Cobble name, logo, palette and typography. Agent terminal sessions start in your login shell, and the Backlog page pages by top-level issue.

    Added

    • The hub and the terminal UI carry the Cobble name, logo, palette and typography
    • A follow-up in an Inbox conversation amends the pending create proposal, and you can restore the proposal it replaced

    Changed

    • Stored theme choices reset once to Cobble on upgrade, and the former palette stays available as the Trau theme
    • Agent terminal sessions start in your login shell so your own wrappers apply, and a slow start shows where the time went
    • The Backlog page counts and pages top-level issues, so sub-issues no longer take page slots

    Fixed

    • The diff of a finished run shows only its own commits, and an interrupted worktree removal no longer leaves a partial tree

  4. v2.71.0

    A Terminal sidebar and a one-line Windows installer

    The Terminal page groups its sessions in a sidebar, names each one from its window title and marks the active ones. Native Windows installs with one PowerShell line.

    Added

    • The Terminal page groups sessions by kind in a sidebar, with a launcher for each agent, the shell and each named command
    • A session takes its name from the window title its process sets, and you can rename it
    • A pulsing dot marks each session that writes output, and each group counts its active sessions
    • A one-line PowerShell installer sets up trau on native Windows and verifies the download

    Changed

    • The sidebar filters sessions as you type, number keys jump to a session and a running session asks before it closes

    Fixed

    • A second update on Windows no longer fails while an earlier build is still in use

  5. v2.70.0

    Evidence checks for checklists and takeover in the hub

    An agent checks a checklist item and proposes a tick with its evidence for you to confirm. The App page sets up a JavaScript workspace monorepo by itself, and takeover opens in a terminal inside the hub on every operating system.

    Added

    • An evidence check on a checklist item has an agent propose a tick with its evidence for you to confirm, and an open item becomes a ticket in one click
    • The App page sets up a JavaScript workspace monorepo by itself, with one service per app, settings from example files, Save and start and Fix with AI
    • Takeover opens a run in a terminal inside the hub on every operating system, and a finished Research Report offers next steps

    Changed

    • The agent draws on the Interview board without asking first, places stencils from the wireframe kits and shows each drawing in the conversation
    • Readiness icons and split advice read the same before and after a ticket is created, and an Interview loads only the related tickets that matter

    Fixed

    • Backlog sub-issues stay under their epic at any depth, a pull request no longer picks up base branch changes and Ghost text works with more models

  6. v2.69.0

    A drawing board for Interviews and a deeper Data page

    Sketch an idea on a drawing board and hand it to the agent in any Interview. The Data page gains column control, filters, search and export, and every ticket now runs in its own worktree by default.

    Added

    • A drawing board in every Interview holds sketches, wireframe kits, images and comments, and the agent reads it and draws on it once you allow it
    • The Data page gains column control, pinned columns, filters, multi-column sort, search across every table, column profiles and CSV or JSON Lines export
    • The Issue context panel lists what a Prepare session added: related tickets, references, searches, webpages and skills
    • A ticket ready for testing gets a How to test section in its description, and Assistant conversations can be deleted in any state

    Changed

    • Each ticket runs in its own worktree by default, so your checkout stays untouched and ready tickets run side by side; a project set to one checkout keeps it

    Fixed

    • A forge outage during a push pauses the run instead of starting a repair, and a local merge lands while the project's app is running

  7. v2.68.0

    An Assistant in the hub and local apps on the App page

    The hub gains an Assistant that answers questions about your runs and steps in when one stalls or fails. The App page runs a project's local app with the services it needs, and Claude Sonnet 5.5 is a supported model.

    Added

    • An Assistant on every hub page answers questions about your runs, takes reversible actions by itself and proposes the rest for you to apply
    • The Assistant opens a conversation when a run stalls or halts, and a paused or quarantined run offers Ask the Assistant
    • The App page starts, stops and restarts a project's local app with its preparation steps, Docker Compose setup and database and cache services
    • A Research Report drafts a Proposed ADR that you review before it is saved, and Claude Sonnet 5.5 is a supported model with its own rates

    Changed

    • Interviews pre-select the recommended answer, send only the answers you chose and fall back one model tier when a default model is unavailable
    • A stacked epic starts a child as soon as the sibling it depends on is green, and the away recap shows each ticket's title next to its ID

  8. v2.67.0

    Sign in to your trackers and inspect your data

    Connect Linear, Jira, Bitbucket and GitHub with a sign-in instead of a pasted token, and browse a project's database on the new Data page. Atlas flows, a starter catalog and a parsed agent feed round out the release.

    Added

    • Sign in to Linear, Jira, Bitbucket and GitHub under Settings → Connections or with trau connect, so a project no longer needs a pasted API token
    • A Data page inspects a project's SQLite, PostgreSQL or MySQL database read-only, runs SQL and answers questions asked in plain language
    • Atlas opens each View full screen with search, export and a version diff, and a typed prompt becomes a narrated flow you can play and regenerate
    • New projects start from a catalog of framework and CMS starters, and operator agents gain tools for run recovery, evidence, secrets and checklists

    Changed

    • Runs show the agent's activity as steps for every provider, dock the steer composer on a live run and name where code went: shipped, merged or in review
    • Interviews can file an epic into a sibling project of the same Workspace, and trau config list documents every setting without a running hub

  9. v2.66.0

    Empty diffs settle as no change and trackers retry

    A run whose build and verify pass without changing code now settles as no change instead of halting the Queue. Tracker calls ride out short outages, and Claude Opus 5.5 is a supported model.

    Added

    • A run that passes build and verify with an empty diff settles as no change: nothing is committed, pushed or opened, and the ticket waits for you
    • Claude Opus 5.5 is a supported model with its own rates, and past Opus 5.5 and Fable 5.1 spend on Costs is corrected to the real prices

    Changed

    • The Queue keeps draining past a no-change run and shows it with its own badge on the Queue, the run pages and in the terminal
    • Linear, Jira and Azure DevOps calls retry after a short outage, and a retried create never files a duplicate issue or comment
    • The test gate reports skipped instead of passed when a change has no test targets

  10. v2.65.0

    Settle ADR conflicts and compare model lineups

    Trau now lets you settle an ADR conflict without leaving the app and brings every model setting onto one Models page. Updates no longer touch your live database while they check a new build.

    Added

    • Settle an ADR conflict from the run page, the Queue or trau adr: authorize superseding the decision, or keep it with a note the repair works from
    • A Models page gathers models by phase and by ticket complexity, with one comparison table of the lineup each complexity band runs under
    • Research and Interviews on a Jira project can read related Teamwork Graph context when you turn it on for that project
    • Build, verify and interview prompts carry a shared font policy for design work, which you can override per project or for every project

    Changed

    • Prepare lists only tickets with an Interview or Draft, and the Queue's running card shows the agent's activities instead of step names

    Fixed

    • trau update and the hub preflight no longer migrate your live database while checking a new build, so a failed update cannot lock the hub out