Repository navigation
What's New
Tag 0.4 lets you decide what each Tag may do without asking, and adds Tags that
work on a schedule, remember what you tell them, and ask each other for help.
Tag.app now runs on Linux too, and adding a Tag to a workspace you already use
takes one step.
Already using Tag 0.3? You don't need to set anything up again. Run
tag upgrade, or let Tag.app update itself. Your Tags keep their settings. See
Upgrade from 0.3.
Highlights
- Permissions per Tag. Each Tag's Details shows its agent's own
permission settings: Codex's sandbox, approval policy and approvals reviewer,
or Claude's permission mode. Rules allow or block a command or tool for
one Tag or for every Tag. A block anywhere wins. - Routines. A Tag can work on a schedule without anyone asking, such as a
weekly digest posted in a channel. Each routine has spend, turn and time
limits and asks its owner before risky actions. See
Routines. - Memory you control. Ask Tag to remember, change or forget a fact for one
channel or for all of them. Each reply states exactly what changed. - Tags that work together. One Tag can ask other trusted Tags for help and
combine their answers in your thread. - Queue or Steer. Message a Tag while it's still working: queue the new
message, or stop the current request and answer the new one next. - Thinking in Slack. Agent activity shows a short summary each time the
model thinks, for Codex and Claude.
Get Tag
Download Tag from the
GitHub release for your
computer:
| Platform | Download |
|---|---|
| macOS (Apple silicon and Intel) | Universal .dmg |
| Linux (Ubuntu, Debian) | .deb |
| Linux (other distributions) | .AppImage |
Windows: coming soon. Until then, use the terminal installer. See the
installation guide.
The install now downloads the 560 MB memory search model in the background, so
you can start using Tag sooner. You can also skip it and use your own
OpenAI-compatible endpoint instead.
Everything new in 0.4
Tag.app
- On Home, Add Tag on a workspace's card adds a Tag to that workspace in
one step: name, description and picture.tag add --workspace T0ABC123does
the same in the terminal. - Routines in each Tag lists its scheduled work, with pause, resume and
run history. - Settings has a new AI tab with AI connections and Memory, which shows
whether memory runs and which search model it uses. - Choosing a model is immediate. Tag saves each account's model list and
updates it in the background. Refresh list reads it again. - Home is quieter with one or two Tags.
Setting up a Tag
- For a Slack organization, setup lists its workspaces and picks the only one
for you. - When Slack's sign-in line expires, Get a new sign-in line gets a fresh
one without restarting setup. - A new Tag that isn't in any channel yet starts, so you can invite it
afterwards.
Permissions and rules
- Codex: sandbox (Read only, Workspace write, Full access), approval policy and
approvals reviewer. Claude: permission mode (Default, Accept edits, Auto,
Don't ask, Bypass permissions). - Always allow in Slack saves a Codex rule for that Tag only.
- In the terminal:
tag NAME settings ai permissionsand
tag settings ai rules.
In Slack
- Saved facts load before every request in their channel. Tag turns off Claude
auto memory and Codex memories for its runs, so everything it remembers can
be listed, corrected and forgotten. - Other apps' bots, such as Hover, can ask Tag to run a skill once you list
them inSLACK_ALLOWED_BOT_IDS. - A channel can search more than its own Slack history with
MFS_CHANNEL_SCOPES. - Tag reads all of a long text file attached in Slack, not only the first
12,000 characters. - With Claude, large images, screenshots and PDF pages up to 64 MB work.
Memory
- Choose the built-in search model or any OpenAI-compatible endpoint (OpenAI,
Ollama, LM Studio, vLLM) under Settings → AI → Memory or with
tag memory embedding. tag autostartkeeps memory running and restarts it if it stops.
See the changelog for every change and fix.
Upgrade from 0.3
Run tag upgrade, or let Tag.app update itself. Nothing needs to be set up
again.
| Change | What happens |
|---|---|
| Permissions | Existing Tags keep the permissions they had. |
| Codex rules | Rules saved with Always allow before 0.4 stay in your global Codex rules. New ones are saved for the Tag only. |
| Agent memory | Tag turns off Claude auto memory and Codex memories for its own runs. Your own use of Codex and Claude Code is unchanged. |
| Thinking | Thinking summaries show in Slack threads. Run tag NAME config set OPENTAG_SHOW_THINKING 0 to keep them private. |
| Handoffs and bots | Handoffs stay off until an owner lists trusted Tags in OPENTAG_PEER_TAGS. Bot mentions stay ignored until a bot is listed in SLACK_ALLOWED_BOT_IDS. |
Known limitations
- Tag.app for Windows is attached to releases but is not yet supported. Use
the terminal installer on Windows. - Codex Block rules need Codex 0.162 or newer. With an older Codex, a Tag
whose rules block or ask stops with a message to update Codex. - Usage budgets are advisory. They don't stop tasks. Routines have their own
per-run limits. - Tag runs with your account's files, credentials, tools and permissions. Use
it with trusted people and workspaces. It is not a hardened multi-tenant
security boundary.
See the release contract, the
installation guide, supported capabilities
and the security policy.