Adjutant reads a project the way a chief of staff reads the file before the meeting: the description, every master document, the notes from your last threads, the knowledge graph and the files you added this week.
Then it puts three or four specific next moves above your chat input,
each with the mode it should run in and one line on why now.
Click one and the prompt is in the box, ready to edit.
Long projects fail in a boring way: too much state ends up in your head. You return after a week – or reach turn ten – and have to remember what is settled, what is open, and how to phrase the next prompt well enough to make the AIs deliver instead of discuss. When you are tired, you ask the safe follow-up or close the tab.
Adjutant takes that cognitive load. It tracks decisions and unfinished work across the project, then proposes three or four ready-to-run prompts with the right mode for each job. If something important stays unresolved, it keeps bringing it back until you finish it or dismiss it. You still choose and send. Adjutant keeps the work moving.
Illustrative example, on the same acquisition project the demo above works through. The brief is where the project stands. Each card is one move, with the mode it should run in and the reason it is on the table today.
The Skybridge acquisition has moved from “should we” to “on what terms.” The retention memo is in a master document and the deal team accepted the pause. The technology valuation Gemini asked for in the first thread is still open, and the partnership route Grok raised has not been worked through.
Every card works the same way. A title, a mode badge, the full prompt and one line of rationale. Click it and the prompt fills the chat input, the mode switches to match, and you edit or send.
Three seconds after the project opens, the preset prompts fade into a loader, and a short brief plus three or four cards take their place. Start typing before the cards arrive and the run cancels.
The brief is two or three sentences on where the project stands. It is not a description of the cards. It is the strategist showing you it read the file.
“Show different ideas” regenerates the set. Twice per state change, then it waits for the project to move.
When the last AI in a Sequential chain finishes, or the Super Mind synthesis lands, or the Debate rounds close, three new cards appear above the input. No brief. You are already in the conversation.
Best for turn six of a long thread, when the useful next question stopped being obvious. Dismiss a card and it stays gone for that turn. Type your own prompt and the cards clear on send.
Cards never appear inside Decision Validation Engine or Research Symphony threads. Those have their own endings.
Click the Adjutant icon in the chat input. The panel opens with two choices. Project runs the same read of the project and shows the brief and cards inside the panel. Paste is the Prompt Adjutant you already use, unchanged: paste a brain dump, get a structured prompt.
Best for consulting the strategist mid-session without waiting for a turn to finish.
On a project that has not met the minimum (one completed thread, a 500-word description), the Project button is there but disabled, with the reason in a tooltip.
Adjutant never recommends from your last message alone. It rebuilds the current state of the project from Master Documents, Scribe notes from recent threads, the project knowledge graph, recent files and the project description. Mid-thread, it also reads the latest turns and the active mode, so the next card can respond to what the AIs just discovered.
It weighs committed decisions above discussion and newer evidence above stale context. A Master Document stays settled unless a newer thread or file gives it a reason to reopen the issue. Older conversation history can compress into summaries; the decisions that define the project do not disappear.
Explore the interactive product demo below. Follow a Sequential discussion, then see how the reasoning carries into Scribe and the Master Document. Those notes and documents are exactly what Adjutant reads before it puts the next move on the table.
Every run follows the same five steps, whether it fires on project open, after a full turn, or from the panel. What changes is the context it reads and where the cards render.
The one constant is the last step. Nothing runs, sends or changes until you click a card and hit send. Adjutant is a strategist with a memo, not an agent with a mandate.
Adjutant proposes. It never starts a thread, never sends a prompt and never changes a decision. The output is always three or four options on the table. The user decides.
Generic advice fails the test. “Analyze your competitors,” “review your pricing” and “validate your assumptions” fit any project. Every card must name the unfinished work, the evidence or decision behind it, and why it matters now. If Adjutant cannot do that, it shows fewer cards.
That is why it can be persistent without becoming noise. An unresolved pricing decision can keep returning until you finish it; dismiss the direction twice and it stops. On thin projects, Adjutant stays conservative rather than inventing context.
What passes the bar
“Stress-test the 78% retention figure. The pause rests on one number from one deck, and Perplexity flagged the cohort data as unverified in turn three.”
A named number. A named thread. A named model. A reason that only makes sense in this project, this week.
Every suggestion carries exactly one mode. The choice follows the shape of the work, not a rotation.
A plan, a number, a draft you produced. Attack it before it moves.
Three threads and a master document that have never been read as one.
Reserved for foundations. Never used for tactical work.
Fresh data from Perplexity, a careful rewrite from Claude. Named on the card, with the reason.
Click a card and the mode switches to the one on the badge. You can switch again mid-thread and the AIs carry the full context across. Your plan determines which modes are available.
The AIs respond in order. Each reads the responses before its own and can challenge, extend or refine the reasoning.
Adjutant picks it when:
The work builds in layers and each AI should read the one before it
The AIs answer in parallel, then a synthesis brings their contributions together. Use a follow-up to explore a disagreement or develop an idea.
Adjutant picks it when:
You have accumulated material that needs pulling together into one answer
Put competing positions into a structured argument. Examine the case for each, the objections and what would change the conclusion.
Adjutant picks it when:
Two defensible positions exist and neither has been argued against the other
Ask the AIs to look for ways your plan could fail. Bring the weak assumptions and practical risks into the discussion before you commit.
Adjutant picks it when:
You produced something (a plan, a number, a draft) that should be attacked before it moves
Work through a research question with stages for gathering sources, analysis, checking, challenge and synthesis. Available on Enterprise.
Adjutant and this mode:
Cards never appear inside a Research Symphony thread. The pipeline owns its own ending.
Take the question back to its underlying assumptions. Have the AIs examine what must be true and rebuild the approach from there.
Adjutant picks it when:
A foundational assumption is in question. Never for tactical work.
An @mention card names the AI it wants and why. See the plan comparison for model access, available modes and usage limits.
Labeled Prompt Assistant in the app. You have a rough idea. Paste it, get a structured prompt. It works on the prompt you already have. Pro and up.
Your hands-on project strategist and a best friend when you are running a long & complex thread. It reads the project and proposes the next three. Frontier, Power and Enterprise, coming soon.
The five AIs have answered and disagreed. It reads the divergence and writes the decision brief. Pro and up.
On a real project they run in order. Adjutant proposes the thread. Prompt Adjutant tightens the wording. The five AIs argue. The Adjudicator settles it. Scribe writes it down. The next Adjutant brief starts from what Scribe wrote.
Adjutant maintains a readable Project Understanding: current position, decisions, open questions, cross-thread patterns and work that has gone dormant. It rewrites as the project changes, so reopening it does not mean rebuilding the mental model from scratch.
You can inspect or regenerate it and audit every card as accepted, edited, dismissed, ignored or pending. Per-project controls let you turn cards off, use Lean mode, disable in-thread suggestions, control auto-triggering and choose three or four cards.
Last updated 2 days ago. 12 threads analyzed. 3 master docs referenced.
Current Strategic Position
The Skybridge project has settled whether to proceed at the current terms and is now working on alternative structures. Commercial clarity is ahead of technical due diligence.
Decisions Made
Pause the acquisition until retention is validated against cohort data. Value the product separately from the customer base.
Open Questions
Whether a partnership captures the technology without the customer risk. Who owns the retention verification, and by when.
Dormant Areas
The regulatory review thread has not been touched in 31 days.
Show Adjutant cards
Master switch. Off disables the strategist on every surface.
Lean mode
Hides Adjutant cards and falls back to the preset prompts.
Show in-thread cards
Cards appear above the chat input after every full turn.
Auto-trigger on project open
Pre-thread cards fire three seconds after you open the project.
Card count
Three or four suggestions per generation.
Four projects, four briefs
The same strategist, reading four different projects. Each panel shows the brief it wrote and the first cards it put on the table.
Strategy Consultants
A client engagement with three completed threads, one memo in a master document and a deck the client sent yesterday. Adjutant reads all of it and points at the two things the memo left open.
Brief
The Skybridge memo is in a master document and the client accepted the pause. The technology valuation is still open. The partnership route has been raised but never argued.
Founders & Operators
A pricing and launch project where the pricing thread closed last week and the onboarding thread never met it. Adjutant notices the two threads have not been reconciled.
Brief
Pricing landed at two tiers last week. The launch checklist has no risk pass yet, and the support-load estimate from the onboarding thread was never reconciled with the pricing thread.
Researchers
A literature review with fourteen sources in the knowledge graph and two threads that disagree. Adjutant finds the disagreement the draft is quietly citing both ways.
Brief
Fourteen sources are in the knowledge graph. Two threads disagree on whether the 2024 replication counts as a failure. No thread has settled it, and the draft cites it both ways.
Investment Analysts
A thesis project where the bull case has a master document and the bear case lives in one thread that ended without a conclusion. Adjutant treats the master document as settled and goes after what is not.
Brief
The bull case has a master document. The bear case lives in one thread from three weeks ago that ended without a conclusion. The margin assumption has not been questioned since.
Adjutant is not a noisy copilot firing advice after every click. It runs when project state can actually change – on project open, after a completed thread, a new file, a Master Document or a description edit. In-thread cards wait for the full multi-AI turn, so they react to finished reasoning rather than interrupt it.
That cadence protects signal quality first and budget second. We would rather call the strongest model less often than weaken the model, strip context or lower the quality of the guidance. Cards persist until the state changes; typing cancels a pending run; Lean mode turns it off.
I kept Adjutant on my own projects for two months because Suprmind was the hardest test I could give it. One early brief caught two threads defining the same activation metric differently and proposed a Debate to reconcile them. I clicked it. That thread produced the definition we still use.
That is when it stopped feeling like another AI feature. It remembers what is unfinished and keeps pushing until the work is done or you tell it to stop. The remaining work is cost tuning – not weaker models, less context or softer guidance. Then it ships to Frontier, Power and Enterprise.
Radomir Basta, founder
“5 AIs were a go-to resource in setting up our new business venture in NYC. From red teaming the initial idea (with harsh feedback), studio market and competitors analysis, to day to day brainstorming about launch phases and website setup. Being able to bounce any idea off 5 AIs, get a clear filtered answer and a todo list in 10 minutes helps a lot.”
CEO, OFF Studio NYC & Funduck Production
“I started using it for competitor research and it just kept expanding – new markets, risk reviews, compliance docs. Five different angles on the same question catches things I would have missed.”
CEO & Co-founder, Miss Amara
“We run everything through Suprmind now – new business ideas, client contracts, marketing strategies. Having five AIs push back on each other in one thread replaced hours of second-guessing between tools.”
Co-founder & COO, Global Digital Marketing Agency
“For analyzing business plans and evaluating client processes, the depth you get from five models reading each other is genuinely different. The Master Document export with custom prompt alone saves me hours on final reports.”
Senior International Adviser, EBRD – European Bank for Reconstruction and Development
Adjutant ships on Frontier, Power and Enterprise. Frontier includes early access to new features, which is where it lands first. Spark and Pro keep the preset prompts and the Prompt Adjutant.
Frontier $95 a month. Power $195 a month. Enterprise on a call. The 7-day trial on Spark, no credit card, shows you the rest of the platform first.
Before you ask
A project-aware strategist. When you open a project, and after every full turn inside one, it reads the project state and proposes three or four specific next moves as cards. Each card carries a mode, the full prompt and a one-line reason. Clicking a card fills your chat input. Nothing is sent until you send it.
No. Prompt Adjutant takes a prompt you already have and restructures it. Adjutant decides which prompt you should write next by reading the whole project. In the app they share one panel: the Project button runs Adjutant, the Paste button runs Prompt Adjutant.
Never. When you click a card it fills the chat input and switches the mode. You can edit, send or dismiss. It does not run a thread, generate a document or change a decision on its own.
Four reasons. The project has no completed thread yet, or its description is under 500 words, so the preset prompts stay. Lean mode or the master switch is off in Adjutant settings. You are inside a Decision Validation Engine or Research Symphony thread, where cards never appear. Or a run failed, in which case the preset prompts stay and nothing else changes.
One mode per card, chosen by the job. Something you produced that needs attacking gets Red Team. Two defensible positions get Debate. Material that needs pulling together gets Super Mind. A questioned assumption gets First Principles. A task that needs one AI’s particular strength gets an @mention with that AI named. Everything else gets Sequential.
Frontier, Power and Enterprise. It is in private testing on the founder’s account today and marked coming soon on those plans. It is not planned for Spark or Pro, because the quality of the suggestions depends on reading a full project with a top-tier model, and a generic suggestion is worse than none. See the plan comparison.
Yes. The Project Understanding tab in Adjutant settings shows the document it maintains: current position, decisions made, open questions, patterns, active and dormant areas, with a Regenerate button. The Recent Recommendations tab lists every card it has shown you, filtered by what you did with it and searchable by prompt text.
It is designed not to. One run per project per session, cards kept until the project changes, two reruns per state change, in-thread cards only after a full turn completes, and a cancel if you start typing. Lean mode switches it off entirely.
Disagreement is the feature.
Adjutant reads the file. You give the order.
