Ten days that changed how the system operates

The Chief of Staff became more reliable

Automatic mode now always starts with the Chief of Staff itself. It answers using company context, memory, recent conversations and, when needed, information from Calendar, Gmail and Drive.

If it understands that the topic deserves deeper work, whether Board, Research, Scenario, Review or a Counselor Reading, it answers first and then asks whether you want to escalate. No heavy workflow is called without your permission.

You can still choose Board, Research, Scenario or Action directly when you already know exactly what you want.

Memory became genuinely alive

bizself.ai now keeps a Living Context: a stable portrait of the company combined with everything that happened recently. Conversations, decisions, tasks, dumps and important changes stop being loose information and continuously feed the system's understanding.

We also added a temporal memory graph. In practice, the system understands not only that a piece of information exists, but what its current state is and what changed over time.

The old memory approval queue was removed. Memory is now written immediately, always with identified origin and one-gesture removal. Only relevant conflicts or exceptions come back for you to decide. Memory needs to accumulate, and it stops growing when every item depends on manual approval.

The Briefing became a management routine

The Morning Briefing is now part of a complete routine:

• update the Living Context
• prepare the daily briefing
• run a voice alignment with the Chief of Staff

Open tasks appear in the briefing with owner, due date and status. The goal is not limited to showing a list. It is to run a short stand-up: what changed, what is late, who owns it and what needs to be updated.

This routine can also run automatically at the configured time.

Tasks are now a single source

Everything you ask the Chief of Staff to track appears in /tasks. If you say:

the system can create, assign, update or complete the existing task. The same task feeds the briefing. Gone is the hidden list inside briefing text living alongside a different list in the tasks area.

You can talk by voice while you navigate

We created Quick Talk, a discreet button near your name. You choose who you want to speak with, press, and start a voice conversation. No other screen to open, no context to choose, no captions to follow. The context is already present.

You can keep looking at a briefing, a decision, a source or a meeting while you talk. The call stays active as you navigate, and recording consent is requested only the first time.

Chat and voice are now one conversation

Text and voice now share the same continuity per interlocutor. Automatic context is always on and, if you want, you can add a specific focus: a memory, source, file, attachment or dictation.

The chat interface also improved a lot: a larger field, clearer controls, history collapsed by default, and the same layout in text and voice modes.

Calendar became more useful before and after the meeting

Smart Calendar now brings together upcoming meetings, recent meetings, history, preparation, transcription and outcome extraction.

Before the meeting, the Chief of Staff prepares a briefing using calendar, company context, memory, recent dumps and, when available, participant emails. Afterward, you can attach the transcript and extract facts, tasks and stakeholder notes.

Preparation can be updated without creating several disconnected versions. It stays attached to the same meeting.

Board, Research and Scenarios became clearer

The AI Board now has its own composition area, where the personas that join the debate live.

Research compares answers from different models and shows explicitly where they agree, where they disagree and what was stated by only one model.

Scenario stress-tests a decision against adversarial futures for market, competition and team, delivering a robustness score and the main risks.

Inside chat, we added an explanation of what each conversation type does, who participates and how deep the work is.

Results can now leave the system

Decisions, debates, research, briefings, meeting preparations, tasks and other outputs can be exported as Markdown, PDF or directly to Google Drive.

You can also configure the company visual identity. The CEO can upload a brand reference or point to a website, review colors, fonts and logo, and apply that identity to generated PDFs.

More control over budget and models

The budget cockpit now shows spend in local currency, by usage type and by period. There are two independent limits:

• recurring limit, daily or monthly
• total cumulative limit

You can also view free periods: today, last 7, 30 or 90 days, year or full history.

For those who use their own key, BYOK is integrated into configuration and Calibration. The model chosen by the CEO is then used across the whole instance.

Onboarding and Calibration were reorganized

The flow now follows a clear sequence: initial interview, review and activation by the administrator, first access, mandatory Calibration and full product release.

The goal is to make sure the CEO never starts with an empty chatbot. They enter a system that already knows their company, their limits, their language and their way of deciding.

Quality and stability improvements

We also made a series of less visible but important improvements:

• more realistic wait times in AI operations
• better history preservation
• larger context for models with long windows
• fixes in the use of configured models
• English improvements
• dictation that only cleans speech, without trying to answer on its own
• meditation audio with clearer voice, consistent volume and no cutoffs
• more stability in automatic Living Context and Briefing generation