Repository navigation
Releases: oye-collective/actvt-macos-releases
Release list
Actvt v1.4.1
Improved
• An agent writing to one session no longer redraws every row of the session list, so Actvt uses less CPU while your agents work with the window open
Fixed
• A loop agent whose session has ended now reads "Ended" in the Overview's Loops card and in actvt-cli loops, instead of staying "Overdue" for ever. Ended loops drop out after a day
• When Actvt's built-in sensitive-data rules improve, sessions counted with the older rules are re-scanned, so the Overview's sensitive-data figures and filters reflect the new rules without waiting for each transcript to change
Actvt v1.4.0
New
• A session's cost and usage now stay in your dashboards and spend tools after Claude Code deletes its transcript (by default after 30 days), so 90 days and All time keep counting them. The deleted text itself is no longer kept or searchable
• With AI PII detection on, the Sensitive Data card shows how much of a long session the AI scan has covered ("AI scan: N of M turns")
Improved
• With AI PII detection on, opening a long session no longer runs the model over the whole transcript at 3–4 cores for minutes. Actvt now scans the turns you're reading, as you read them, saves what it has scanned, and stops when you leave the session
• Scrolling the session list and transcripts is much smoother: the once-a-second metrics update no longer redraws the agent views, and session rows are lighter to draw
Fixed
• A clock that jumped forward once, then was corrected, no longer locks a paid licence while you're online. The next check with our server resets it
Actvt v1.3.0
New
• Loops card on the Overview (Premium): every loop agent on this Mac and on your Remote Sessions hosts, with what it is doing (working, asleep until a time, or overdue) and what its last tick said. An overdue loop is counted in the header and shown first. Click a row to open its session. Remote hosts need actvt-cli 0.7.0
• actvt-cli loops: lists each loop agent's state from its own schedule, for scripts and watchdogs (actvt-cli loops --json)
Improved
• With AI PII detection on, Actvt launches lighter: the model now loads the first time you open a session, not at every launch
• In new and updated sessions, on-device sensitive-data detection now catches secrets in the forms transcripts actually hold: .env lines, JSON keys, Bearer tokens, private keys, database URLs with passwords, and GitLab, Hugging Face and npm tokens
Fixed
• If your first launch was offline, or our server was briefly unreachable, you never got your free trial. Actvt now keeps asking until the server answers
• Claude Code no longer loses its connection to Actvt's MCP server when the token changes, while Settings still said connected
• When port 4096 was busy at launch, Actvt moved its MCP server to another port and every connected agent lost it. It now waits for 4096 to free up
• The hotkey's second key is no longer also typed into the app in front
• Remote agent notifications no longer arrive twice
• Sessions are dated by their last message, so copied or restored transcripts no longer all look recent
• The agent_usage_limits MCP tool now names each usage window by its length, so a weekly window is no longer called "session"
Actvt v1.2.8
Improved
• Actvt uses about half the CPU it did while an agent is writing to a long session. It no longer re-reads the whole transcript after every line the agent adds
• With AI PII detection on, Actvt starts lighter: it no longer re-checks its 1.6 GB model from scratch at every launch
Fixed
• With AI PII detection on, if the on-device model ever fails to run, Actvt now says so in the session's Sensitive Data card ("AI scan failed: names not masked") instead of quietly showing only the rule-based matches. A failed scan is no longer remembered as the session's result, so the next time you open the session it is scanned again. Sessions scanned by earlier versions are rescanned once
• With AI PII detection on, Actvt no longer reports "insufficient disk space" for a model it has already downloaded
• Actvt's window now keeps its size and position after you restart your Mac
Actvt v1.2.7
New
• Agent sessions now show memory: what a Claude Code agent knew, opened and saved in its memory during a session. The inspector counts them, the transcript marks the moment a memory was opened or saved, and Show memories lists each one with when it happened and what changed. Click any step to jump to it. For sessions on remote hosts, this needs actvt-cli 0.6.1
Improved
• With semantic search on, Actvt no longer keeps up to two CPU cores busy while an agent is working. It now updates a session's search once the agent pauses, instead of after every line it writes
• Actvt no longer refreshes its session list every 15 seconds when it was started in the background with no window open, as it does at login
Fixed
• An agent running on a loop no longer shows up as gone quiet, or as waiting for your input, while it sleeps between runs. Actvt still tells you if it doesn't wake up when it said it would. For agents on remote hosts, this needs actvt-cli 0.6.1
• actvt-cli usage could crash on exit on Linux and print nothing. Re-run the install command on remote hosts to get actvt-cli 0.6.1
• actvt-cli usage can now read Claude plan usage on a host where Claude Code signs in with a setup token (CLAUDE_CODE_OAUTH_TOKEN), as headless agent hosts usually do. It reported "needs sign-in" there before
Actvt v1.2.6
New
• actvt-cli usage --json reports the Claude and Codex plan usage of the machine it runs on: session and weekly limits with their reset times, on Mac or Linux. Re-run the install command on remote hosts to get actvt-cli 0.6.0
Improved
• Actvt uses about a third less CPU in the background. It no longer starts a process scan every 15 seconds to see which agents are running
• With Actvt in the background, sessions show an agent starting or stopping within about 10 seconds
Fixed
• The default horse icon in the menu bar showed extra legs. It runs on four again
Actvt v1.2.5
Improved
• Actvt uses about a fifth less CPU while its window is open
Fixed
• Session costs left out what subagents spent, by up to 30 % on sessions that used them. Past sessions are corrected after the update
• An expired licence was re-checked with the server on every network change, about once a minute
• Two agents calling Actvt's MCP server at the same time could receive each other's answers, or none
Actvt v1.2.4
New
• Ports an agent left open are traced back to the session that started them, and you're told when a session ends with one still running
• CPU use is attributed to each agent session
• Spend is counted on the day it happened, not the day a session last ran
Improved
• The menu bar icon and system sampling use far less CPU
• Meaning search now covers the whole of long sessions, not just their start
• Clicking a remote notification opens its session. Re-run the install command on remote hosts to get actvt-cli 0.5.0
Fixed
• Codex sessions from the ChatGPT app showed no messages, and edits were missing from files touched
• Sessions interrupted by an API error showed no cost
• Searching for a hyphenated term, such as a file or branch name
Actvt v1.2.3
New
• Basic licences no longer show an expiry date.
Improved
• Anonymous usage data, when you have it switched on.
Fixed
• The welcome screen no longer runs past the top of its window.
Actvt v1.2.2
New
• Sessions from the pi coding agent, local and remote, with transcripts, cost, search, resume and the MCP tools
• Notifications lead with the session name. The project sits underneath and the body says what happened
• With previews hidden, the lock screen shows only the kind of event, never a session name or transcript text
Improved
• actvt-cli 0.4.0 lists pi sessions. Re-run the install command on remote hosts to pick it up