noHuman Team 4.7: smaller conversations, clearer task progress

Long tasks leave a lot behind: messages, browser results, drafts and decisions. In noHuman Team 4.7, you can choose when a noHuman summarizes that conversation to make room. It also receives its saved open tasks and recent progress reports when it works again.

The release supports macOS and Windows. See the install page for the installers available now, or read what's new in 4.7.0.

Let a quiet conversation make room

Open a noHuman's Settings → Work limits. Compact when idle starts a summary after the conversation has been quiet for the time you choose and has reached the size you set.

The defaults are 10 quiet minutes and 100,000 tokens. The runtime leaves smaller conversations alone and waits while the noHuman has running or queued work. You can change either threshold or turn the setting off.

Compaction replaces older working context with a summary. Your saved chat history, Memory and files remain separate. The summary can miss a detail, so keep exact requirements and important decisions in saved notes or source files. Our guide to context and Memory explains the difference.

Make room after a delegated task

The same settings page has Compact after a delegated task, off by default. Enable it for a noHuman that handles separate assignments and would benefit from a smaller conversation before its next task.

After a completed delegated task, the runtime waits briefly for a follow-up before summarizing. It leaves small conversations alone. For work that needs the exact detail of the last exchange, you can leave this setting off and use idle compaction instead.

Keep track of unfinished work and milestones

A noHuman receives a list of its recent unfinished tasks with each turn. That list includes who delegated the work, the latest report and recorded milestones. The task records stay on disk across restarts and conversation summaries.

You can ask your lead:

Ask the Developer to check the checkout flow. Have them report when they reproduce the problem, finish the fix and complete the tests.

The lead can name those milestones in the delegation. When the worker reports reaching one, the lead gets the update. Progress still depends on the worker reporting what happened; a saved milestone is a checkpoint, rather than proof that the result is correct. You can open the conversation or inspect the output to review it.

Choose the model's reasoning level

Open the model picker beside the chat's Send button. For models that report reasoning choices, Reasoning shows the levels that connection offers. Default leaves the choice to the model's provider.

A change applies from the next message. Higher levels can spend more tokens and take longer, so use them for tasks where that trade-off helps. The available choices vary by model; the picker shows what the running connection reports.

Fewer repeated notes and misplaced questions

The runtime now sends changed working notes with the next message while keeping stable instructions separate. That gives providers a better chance to reuse unchanged context. Token use and savings still depend on the model, its cache and the task.

Questions raised during delegated work also use the same routing across the two question paths. A team lead can answer questions about its own assignment, while decisions that need you return to the Founder. The team's helper model can assess a question's meaning in the language the team uses, with the existing checks as a fallback.

The release also rejects empty placeholder tool requests before they reach the approval queue and fixes the source build order that caused plugins to read outdated shared types.

Browser controls and test scripts are included too

This release also includes the browser and phone-connection changes developed during 4.6. Each noHuman's Capabilities page explains its browser behaviour and lets you switch browser tools off. A noHuman that tests websites can connect its own Playwright scripts to its browser for cases such as another time zone or two signed-in users. Your browser restrictions still apply.

Read the browser controls and phone connection guide for those changes. You can also link up to three phones to the same computer, each with its own keys.

What about the phone companion?

The phone companion remains in preparation for the App Store and Google Play. Its current development build improves voice input: tap to dictate, stop before sending, or hold to speak and send on release. You can change the dictation language from the message bar. Speech recognition uses the phone's speech service, with on-device recognition where that device and language support it.

These mobile changes need the updated mobile app; installing desktop 4.7 does not replace the app on your phone.

Update without resetting your team

Choose Check for Updates in the desktop app or use the install page. Your teams, conversations, Memory and workspace live outside the app bundle and stay saved when you replace it.

Read the 4.7 release notes for the complete list. The monthly subscription includes a 7-day trial; AI provider charges are separate.