LLypses

User Manual

10 chapters 3 appendices Version 0.6.8

← Back to LLypses

Contents
  1. Chapter 1 — Your first five minutes
  2. Chapter 2 — Who is in the room
  3. Chapter 3 — Making sure the board can answer
  4. Chapter 4 — Rooms
  5. Chapter 5 — Giving the board your world
  6. Chapter 6 — Keeping the record
  7. Chapter 7 — Building your own member
  8. Chapter 8 — Strategy into execution
  9. Chapter 9 — What it costs
  10. Chapter 10 — When something is wrong
  11. Appendix A — Every member at a glance
  12. Appendix B — Every built-in room
  13. Appendix C — Which provider to use for what

About this manual#

LLypses is a boardroom of AI members. You are the CEO. You bring a question, and a room of specialists answers it — hearing each other, disagreeing with each other, and building on what came before.

This manual is organised by what you are trying to do, not by which tab a button lives on. Each chapter is one job, start to finish. You do not need to read them in order, but Chapter 1 will save you time whatever else you do.

# Chapter The job
1 Your first five minutes Get one real answer, today
2 Who is in the room Know who to ask, and how
3 Making sure the board can answer Keys, Roll Call, and models
4 Rooms The right people for the question
5 Giving the board your world The Knowledge Base
6 Keeping the record Minutes, exports, and continuing
7 Building your own member Members you invent yourself
8 Strategy into execution The Workshop handoff
9 What it costs Seeing and controlling spend
10 When something is wrong Fixing the common problems
  • Appendix A — Every member, at a glance
  • Appendix B — Every built-in room
  • Appendix C — Which AI provider to use for what

Chapter 1 — Your first five minutes#

The job: get one real answer out of LLypses today, without understanding all of it first.

What LLypses is#

You type a question. Several AI members answer it — each with their own speciality, each aware of what the others just said. @CFO argues cost, @CRO argues risk, @CTO argues feasibility, and you decide.

That is the difference between LLypses and a chat window. You are not talking to one assistant. You are chairing a room.

What LLypses is not#

It does not build software, run your business, or dispatch a team. When a member says a plan is ready, that means the plan is ready — you or someone you hire still does the work. Members produce specifications, analysis, critique and decisions. They do not produce shipped products, and they cannot carry out tasks in the world on your behalf.

If a member ever offers to notify you when something is finished, treat that as a figure of speech. Nothing is running in the background.

What you need#

Three things, and two of them are keys that are easy to confuse.

  1. LLypses installed — the installer from Red Warden Studios.
  2. Your LLypses licence key — proves you bought the software. It arrives in your purchase email and looks like ABCDE-FGHIJ-KLMNO-PQRST: groups of letters and numbers separated by hyphens.
  3. At least one provider API key — this is what actually pays for the thinking. You get it from an AI provider such as Google, OpenAI or Anthropic, on their own website, and it usually starts with letters like sk- or AIza.

⚠️ These two keys are not the same thing#

This is the single most common problem people hit, and it does not look like what it is.

Your licence key unlocks LLypses. It goes in Settings → License, once, and never anywhere else.

Your provider API key pays for each answer. It goes in Settings → API Keys, in the box next to that provider's name.

If you put your licence key into a provider's API key box, LLypses will warn you. Read that warning. If you save anyway, every member using that provider goes quiet, and it will look like the app is broken when it is not.

An easy way to remember which is which: one key, one payment, forever — that is your licence. One key, many small payments as you work — that is your API key.

Why LLypses asks for your own API key#

LLypses does not resell AI usage. You bring your own key, you pay the provider directly at their prices, and nothing you type passes through Red Warden Studios. Your costs are yours to see and control, and your conversations stay between you and the provider you chose.

Most providers offer a free tier that is enough to try everything in this manual.

Step 1 — Install and open#

Run the installer and open LLypses. You will see a short welcome tour. You can replay it at any time from Help → ▶ Replay welcome tour.

Step 2 — Enter your licence key#

Go to Settings → License, paste the key from your purchase email, and activate. This is once per machine.

Step 3 — Add one provider API key#

Go to Settings → API Keys. There is a row for each provider LLypses supports: Google, OpenAI, xAI, Groq, Anthropic, Mistral and Cohere.

Start with one. You do not need all seven, and one is enough to run a full board. Google is the usual first choice — it has a generous free tier and most members default to it.

Paste the key into that provider's box and save.

If a warning appears saying the key looks like your LLypses licence key, stop and check. That warning is almost always right.

Step 4 — Ask who can answer#

Go to the Roster tab and press 📞 Roll Call → Run check.

Roll Call tells you which members can answer right now, and why any of them cannot. It is free, and it sends nothing to any provider — it is reading what you have configured, not spending your quota.

If a member cannot answer, Roll Call names a working replacement model and offers to switch them. It only ever proposes. Nothing on your board changes until you press Apply.

Members belonging to providers you have not added a key for will say so. That is expected, not a fault.

Step 5 — Ask the board something#

Go to the Boardroom tab and type:

@ALL I am thinking about raising my prices by 15%. What am I missing?

Press send, and watch. Each member answers in turn, and each can see what the members before them said. @CFO will talk margin, @CRO will talk about the customers you might lose, @CCO will ask whether you have contracts that forbid it.

That is your first board meeting.

What you should see#

  • Answers arriving one after another rather than all at once. This is deliberate — later members are reading earlier ones.
  • Members disagreeing with each other. Also deliberate.
  • A running total of what the conversation cost, in the Usage tab.

If nothing answers#

Work down this list in order:

  1. Roster → Roll Call → Run check. It will usually name the problem outright.
  2. Check Settings → API Keys. Is your provider key in the provider's box, and your licence key in Settings → License?
  3. Restart LLypses. If you have just changed a key, a restart makes sure everything picks it up.
  4. Still stuck? Use the Feedback tab to send a bug report. It packages up what LLypses knows about the problem. Your keys are never included.

Chapter 2 — Who is in the room#

The job: know who to ask, so you stop addressing everything to everyone.

Three ways to address the room#

By name. Type an @tag and only that member answers.

@CFO What does a 15% price rise do to my margin?

Everyone. @ALL summons the room you are currently in. Each member responds in turn and hears the ones before them.

@ALL Should I take on a business partner?

By clicking. Members in the sidebar are buttons. Click one to light it up, click again to turn it off. Whoever is lit answers your next message, with no tag needed.

Nobody in particular. A message with no tag goes to whoever is currently lit in the sidebar.

⚠️ This changed in version 0.6.6#

Before 0.6.6, an untagged message went to @COO alone. It now goes to every member you have lit, and each one is a separate call billed to your provider key.

Your selection starts as whoever is already in the room you are working in, so nothing is lit that was not already there. But if you light eight members and send an untagged message, eight members answer and you pay for eight.

Typing an @tag still overrides whatever is lit, exactly as before.

You can address several members at once by listing their tags:

@CTO @CISO Is it safe to store customer addresses the way we do now?

The people at the table#

LLypses ships with fifteen members. You do not have to learn them up front, but knowing roughly who exists saves you asking the wrong person.

The executives#

Tag Role Ask them about
@COO Chief Operating Officer Coordinating the room, summarising, deciding what happens next. Answers when you tag nobody.
@CTO Chief Technology Officer Architecture, engineering, whether a technical plan is sound.
@CFO Chief Financial Officer Margin, pricing, budgets, anything with numbers attached.
@CIO Chief Information Officer Deep synthesis of dense or complicated topics.
@CISO Chief Information Security Officer Security, vulnerabilities, protecting customer data.
@CRO Chief Risk Officer What could go wrong, failure modes, the downside case.
@CCO Chief Compliance Officer Rules, regulations, contracts, consistency.
@CGD Chief Graphic Designer Brand, visual direction, creative briefs.
@CODEX Chief of Implementation Writing the actual code once the board has decided what to build.

@CODEX is worth knowing about properly, because its title undersells it. The other members discuss, weigh and decide. @CODEX builds. Hand it a settled decision and it returns complete working functions, scripts, diffs and commit messages — not outlines or scaffolds. It also reads existing code, debugs from an error message or a description of the symptom, and verifies that a change does what it claims.

If a task is underspecified it asks you the single most useful question rather than guessing.

Pair it with @CTO. @CTO decides how a thing should be built; @CODEX builds it. Asking @CODEX to make an architectural decision, or @CTO to write you a finished script, gets a worse answer from both.

The specialists#

Tag Role Ask them about
@RESEARCH Head of Research Anything current — news, prices, live regulations, recent events.
@INTEL Head of Intelligence Long documents, reports, filings and papers, returned as sourced analysis.

@INTEL is not a summariser. Give it a document and ask a question of it — "what are the risks in this contract", "what should I act on" — and you get structured analysis with its sources shown, rather than a shorter version of what you already had.

⚠️ Only @RESEARCH can see the live web. Every other member works from what its model learned during training. If your question depends on something recent — a price, a rule, a headline — route it through @RESEARCH, or you will get a confident answer that is out of date.

The Braintrust#

Four members who show up as people rather than executives. No titles, no corporate register — useful when you want to think out loud rather than receive a briefing.

Tag Who they are Ask them about
@FRANK The Straight Shooter An honest read. What is good, what is not.
@SPARK The Idea Generator Options you have not considered.
@ARCH The Prompt Architect Getting better answers out of AI, including this one.
@CLASH The Devil's Advocate Everything wrong with your idea.

Try them when the boardroom feels too formal for what you are actually wrestling with:

@FRANK @CLASH — gut check: I want to quit my job and do this full time.

Your Inner Board#

Inner Board is your standing board — the members who answer @ALL when you have no other room open. You choose who is in it, at the top of the Roster tab. Everyone you leave out is still listed, so you can always see who is available but sitting out.

Eight members are left out by default: @CGD, @RESEARCH, @INTEL, @CODEX, and the four Braintrust members.

This is about seats, not about them. Every member on your board is one request each time you use @ALL. On a free provider key the limit you hit first is how many requests you may make per minute — so a board of fifteen turns one question into fifteen calls and can rate-limit you on your very first message. Nine seats keeps a default @ALL comfortably inside a free tier.

Nothing is being withheld. Every one of the eight answers its own @tag immediately, and you can tick any of them into your Inner Board whenever you want them in the room. If you are on a paid key and the request limit is not your constraint, add them all.

Two things people get caught by:

  • Creating a new member does not add them to @ALL. They answer their own @tag straight away, but they join the room only when you tick them into the Inner Board or add them to a room.
  • Editing the Inner Board happens in Roster, not in Rooms. The Inner Board is the one room that always exists, so it is edited where members live.

Why they argue#

The members are built to debate and to build on each other, not to agree. When @CRO contradicts @CFO, the software is working. You are chairing a boardroom, not querying a search engine — the disagreement is the product, and deciding is your job.

Chapter 3 — Making sure the board can answer#

The job: stop guessing whether it is working.

Roll Call#

Roster → 📞 Roll Call → Run check.

Roll Call is the answer to "is this thing on?". It reports which members can answer, which cannot, and why. It costs nothing and sends nothing to any provider.

Run it whenever you:

  • have just added or changed an API key
  • see a member stay silent
  • are about to start something that matters

"Not checked" means Roll Call has not looked at that member yet. It is not a fault with your board or your key.

When a member cannot answer, Roll Call names a replacement model that works and offers to switch them. It only proposes. Nothing changes until you press Apply.

Test my keys#

Test my keys is the one action that spends your provider quota. It makes real calls with your own keys to find out what they can genuinely reach — which is the only way to know for certain, because providers do not reliably publish what a given key is allowed to use.

LLypses tells you how many requests it will use before you press it.

Run it once after setting up, and again if a provider changes your plan.

You can check one provider instead of all of them. Underneath the main button is a row with one entry per provider, each showing its own request count. If you have only changed a member onto a different Anthropic model, checking Anthropic answers the question — a full sweep spends requests on keys nothing has changed about.

One check covers every member on that provider, however many different models they are set to, because the question being asked is "what can this key reach" — and that is a property of the key rather than of any one member.

A model shown as unavailable is one LLypses tested and could not reach. Models it has never tested are not flagged at all, because not knowing is not the same as knowing it is broken. If a model you expect to work is marked unavailable, check that provider again before changing anything — a provider having a bad minute during a check looks identical to a model you cannot use.

Free tier or paid tier#

Roll Call detects this automatically for Google and asks you for the others, because most providers give no way to tell from the outside.

It matters because the two run out differently:

  • A free key runs out of requests. You get a fixed number per minute or per day, whatever their size.
  • A paid key runs out of budget. You get as many requests as you are willing to pay for.

LLypses optimises for whichever one applies to you.

Auto — letting LLypses choose the model#

Every member can either run a model you pin, or let LLypses choose. You will find this on each member in the Roster tab, with three settings:

Setting What it does
Off — run the model above Always use the model you picked. Nothing is chosen for you.
Optimized — best at this price or lower Upgrade to a better model as long as it costs no more than the one you chose.
Performance — most capable your key reaches Use the strongest model your key can actually get to, whatever it costs.

Auto only knows what has been measured. If a member's Auto setting appears to do nothing, run Test my keys — Auto cannot choose a model it has never confirmed your key can reach.

A sensible starting point is Optimized for most members and Performance for the one or two you rely on for hard thinking.

Keeping models current#

AI providers release, rename and retire models constantly. If a member starts failing on a model that used to work, that is usually the provider having changed something rather than a fault on your side. Roll Call will spot it and propose a replacement.

Chapter 4 — Rooms#

The job: stop reading fifteen opinions when you needed three.

The idea#

A room is a named group of members with a prompt that pre-fills when you join. Instead of tagging four members by hand every time, you open a room and start typing.

Click 🏛 in the Boardroom to open the room picker. Closing the picker without choosing puts you in your Inner Board.

The rooms you already have#

LLypses ships with eleven. The most useful starting points:

  • 🏛 Inner Board (Default) — your standing board. What @ALL means when no other room is open.
  • 🛋 The Braintrust — Frank, Spark, Arch and Clash as people rather than executives. Bring whatever you are chewing on.
  • 💡 Gut Check — Frank and Clash only. Drop an idea in front of the two who will tell you the truth.
  • 🔥 Devil's Advocate — the full board, pointed at tearing your idea apart.
  • 🧠 The Advisor — boardroom thinking applied to personal decisions and purchases.

The rest are focused working groups: Business Strategy, Tech Review, Risk & Compliance, Research & Intel, Creative Workshop, Financial Review and Creator Table. Appendix B lists all eleven with their members.

Building your own#

The Rooms tab, next to Roster, is where you build rooms of your own.

  1. Give it a name.
  2. Tick the members you want from the list of everyone you have — built-in and your own.
  3. Add an opening prompt that pre-fills when you join.

Members are chosen from checkboxes, not by typing tags, so a room cannot end up missing someone you thought you had added.

A good room is narrow. "Everyone who might have an opinion" is what @ALL already does.

Rooms that earn their keep#

  • Your weekly numbers — @CFO and @CRO, opening prompt "Here are this week's figures: "
  • Before I publish — @CCO and @CISO, opening prompt "Check this for anything that could get me in trouble: "
  • Customer problem — @COO and @CRO, opening prompt "A customer is unhappy about: "

Editing the Inner Board#

The Inner Board is edited in the Roster tab, not in Rooms. It is the one room that always exists.

Chapter 5 — Giving the board your world#

The job: stop explaining your business at the start of every conversation.

What the Knowledge Base does#

The Knowledge Base holds things you want every member to know: how your business works, who your customers are, what you sell, what you have already decided.

Anything marked active is put in front of every member on every turn, automatically. You never have to mention it or attach it.

Putting things in#

The Knowledge Base tab accepts four kinds of thing:

What How
A note Free-form text you type yourself. Best for things that live in your head.
A web page Paste a URL and LLypses reads the page.
A folder Give it a folder and LLypses indexes the text files inside, including subfolders.
Past AI chats Import your history from Google Gemini Takeout or a ChatGPT export.

When you add a folder, LLypses counts what it found and asks before it commits — "431 files across 26 folders will become knowledge entries." Read that number. A folder import cannot be undone in one step; entries have to be deleted afterwards.

Build and dependency folders — node_modules, .git, dist, .venv and their kin — are skipped automatically. They are full of text files that are not knowledge, and indexing them costs you on every turn thereafter. There is also a ceiling on how many entries one import can create; if you reach it, narrow the folder or the file types and import the rest separately.

Turning entries on and off#

Every entry has a badge you click to make it active or inactive. Inactive entries stay saved but are not sent to anyone.

This matters more than it sounds. Active entries are added to every member's context on every turn — so a large entry you no longer need is a cost you pay continuously.

Organising with folders#

Entries can be filed into folders, which collapse and expand from the arrow beside the name. A folder shows how many entries it holds and how many of those are active — useful when the count on the Roll Call card is higher than you expected and you want to know where it is coming from.

To do this Do that
Put entries in a folder Tick them, then 📁 Move to…. Type a new name or pick an existing one.
Take entries out of a folder The same control, with the name left blank.
Rename a folder Hover the folder row and click the pencil. Nothing is deleted.
Delete a folder Hover the folder row and click the bin.

Deleting a folder asks you which you meant, because there are two different answers: Remove folder only keeps every entry and simply ungroups them, and Delete N entries destroys the content. Importing a folder is one click; getting it back is not.

Folders stay open or closed the way you left them, between sessions.

Selecting several at once#

Tick the box on any entry and a row of batch actions appears — delete, activate, deactivate, and move to a folder. Select all at the top right takes everything, and Clear beside the count releases it.

Every folder row has its own tick box, which selects or deselects everything inside that folder. That is the one to reach for when you want to clear out most of a Knowledge Base but keep one folder intact: Select all, then untick the folders you are keeping.

Selecting everything and deactivating is the fastest way to get a bloated Knowledge Base back under control: switch it all off, then turn back on only what the board actually needs. Nothing is lost by deactivating.

What LLypses already knows#

LLypses ships with its own reference material about AI models and providers — what each one offers, how they differ, which are current. It is what stops your members answering with last year's model line-up.

It is part of the application, not part of your Knowledge Base. You will not see it in your list, you cannot edit or delete it, and it updates when LLypses does. That is deliberate: it is machinery, like the instructions each member carries, and a board whose reference material had been switched off would give worse answers with nothing on screen to explain why.

It does cost you something — see below — and Roll Call says how much.

⚠️ If you edited any of that material in an earlier version, it stays yours. LLypses can tell the difference between what it shipped and what you changed, and it never takes back something you wrote. Your edited copy remains in your list, still yours to change or remove.

Seeing what it costs#

Roster → Roll Call reports what your Knowledge Base costs: roughly how many tokens it adds to every member on every turn, which entries are the largest, and it offers to switch off the ones dominating your context.

It also reports what LLypses' own built-in knowledge costs, as a separate line. There is no button beside that one, because there is nothing for you to switch off — but it is counted in the total rather than left out, so the figure matches what your provider actually bills you.

Check this if answers start feeling slow or expensive.

What to put in first#

Start with one note, a page long, covering:

  • what your business does, in a sentence
  • who your customers are
  • how you make money
  • the two or three constraints that are always true (your budget, your team size, the platform you are stuck with)

That single note changes the quality of every answer you get afterwards, far more than any model setting.

⚠️ The Knowledge Base is a record, not a boundary. Members treat it as authoritative on the subjects it covers, but its silence is not a verdict. If something is missing from it, that is a gap in your notes, not evidence the thing does not exist.

Chapter 6 — Keeping the record#

The job: find that thing the board said three weeks ago.

Everything is saved#

Every session is saved automatically as it happens. You do not have to remember to save, and nothing is lost if you close the app.

Browse them in the Minutes tab. Click any session to reload its full transcript.

Naming sessions#

LLypses suggests a name from your first message, which you can accept or replace. A named session is far easier to find in three weeks than Untitled Session.

Getting a session out of LLypses#

Any session can be exported three ways:

Format Best for
⬇ MD Plain text you will paste somewhere else.
⬇ DOCX Sending to someone who expects a document.
⬇ PDF Sharing something that should not be edited.

The file is saved to your Downloads folder, and LLypses shows you the path with a Show in folder button.

Continuing an old session#

▶ Continue session reopens a past conversation and picks up where you left off, with the members remembering what was said.

Use it when a decision comes back around, rather than starting again and re-explaining.

Turning a session into permanent knowledge#

🧠 → KB sends a whole session to the Knowledge Base, so what the board worked out becomes something every member knows in future.

This is the most under-used feature in LLypses. A session where you settled your pricing is worth far more as a permanent fact about your business than as a transcript you have to remember to go and find.

Be selective — everything you send becomes context you pay for on every turn.

Session briefs#

Ask the room for a summary at any point and you will get a structured brief: who contributed what, what the room agreed on, and what is still open.

It is the fastest way to end a long session with something you can act on.

Chapter 7 — Building your own member#

The job: get a specialist the built-in fifteen do not cover.

When to build one#

Build your own member when you keep needing the same kind of thinking and no existing member quite fits — a specialist in your industry, a stand-in for a particular customer, a reviewer with a specific bar.

How#

Roster → Add Member. You give it:

Field What it does
Tag How you address them, e.g. @LAWYER. Keep it short — you will type it constantly.
Title What they are. Shown in the room.
Model Which AI runs them, or leave it on Auto.
Colour How their answers are marked on screen.
System prompt Who they are and how they should think. This is the important one.

They respond to their @tag immediately. Add them to the Inner Board or to a room when you want them in @ALL.

Writing a system prompt that works#

The system prompt is the member. A vague one produces a vague member.

Say what they are, what they are for, and how they should behave when they disagree with you. That last part matters most — a member who agrees with everything is worth nothing.

A workable shape:

You are a commercial property surveyor with twenty years in the UK market.
You assess buildings for structural risk and resale exposure.

Be specific about figures and cite the standard you are applying. When you
do not have enough information, ask the single most useful question rather
than assuming.

When the CEO's plan carries a risk they have not mentioned, say so plainly
and early, even if they did not ask.

Compare that with "You are a helpful property expert", which produces exactly what it describes: something helpful, and nothing else.

Editing the built-in members#

You can override any built-in member's model, colour or system prompt in Roster → The Board. Useful when @CFO should think in your currency and your market rather than in general.

Chapter 8 — Strategy into execution#

The job: get a decision out of the boardroom and into something that actually gets built.

The idea#

The Boardroom is strategy. The Workshop is execution. A workshop brief is the handoff between them — a written record of what was decided, in a form a developer or an AI coding tool can pick up and act on.

Setting it up once#

Settings → Workspace → Projects Folders. Point LLypses at where your projects live. Auto-detect will usually find it.

Without this, the project picker is empty.

You can add more than one folder. If your work is split across several locations, click + Add folder and add each one; every project under any of them appears in the same Target Project list, with nothing to switch between. If the same project name exists in two folders, the one higher in the list wins — so keep the order deliberate, or rename one of them.

Sending a brief#

After the board reaches a decision, click Send to Workshop. Choose the project, and LLypses writes the brief into that project's workshop_queue folder, alongside the files a coding assistant looks for automatically.

Write the Directive yourself. It starts empty on purpose. The Directive is the part a coding assistant acts on, so it needs to be an instruction — "Add directory exclusions to the folder crawler and cap it at 500 entries" — rather than a summary of the discussion. The discussion travels with the brief anyway, underneath, and your last message goes into CEO Notes as framing.

A vague Directive is the single most common reason a brief comes back as the wrong work.

The next time you open that project in a coding tool, the brief is waiting.

What makes a good brief#

Send a brief when a decision is settled, not while it is still being argued. A brief that records an unresolved debate produces work that has to be redone.

Ask the room for a session brief first, confirm it matches what you actually decided, then send it.

Chapter 9 — What it costs#

The job: know what you are spending, and spend less.

Where the money goes#

You pay your AI provider directly, at their prices. LLypses takes nothing per message. What you spend depends on three things:

  1. How many members answer. @ALL with twelve members costs roughly twelve times what one member costs.
  2. Which models they run. The gap between the cheapest and most capable model at the same provider can be twenty-fold or more.
  3. How much context goes with every message. Your Knowledge Base and the conversation so far are sent with each turn.

Seeing it#

The Usage tab shows what you have spent — for the current session and over time, broken down so you can see which members and which models account for it.

Check it after your first serious session. It is usually less than people expect, and it tells you where to aim if it is not.

Six ways to spend less#

  1. Tag individuals instead of @ALL. Most questions need one or two members, not twelve.
  2. Use rooms. A three-member room costs a quarter of a twelve-member board.
  3. Turn off Knowledge Base entries you are not using. Active entries are sent to every member on every turn. Roll Call shows you which are largest.
  4. Set Auto to Optimized rather than Performance for members doing routine work. Optimized never costs more than the model you picked.
  5. Keep sessions focused. A long conversation carries its whole history with every new message. Start a new session when the subject changes.
  6. Use a free tier while you are learning. Most providers offer one, and it is enough for everything in this manual.

Sessions and their history#

A long session gets more expensive as it goes, because each new message carries everything before it. If a conversation has wandered far from where it started, save it and begin a new one — you can always send the old one to the Knowledge Base if the conclusions matter.

Chapter 10 — When something is wrong#

The job: fix it yourself in under five minutes.

Start here, always#

Roster → 📞 Roll Call → Run check.

It is free, it sends nothing to any provider, and it identifies most problems outright. Do this before anything else.

Nobody answers at all#

First, look for an amber banner across the top reading "LLypses has lost its connection to the boardroom." If it is there, that is the answer and the rest of this list does not apply — the part of LLypses that talks to the AI providers has stopped. Press Try again; if that does not clear it, press Restart LLypses. Your session, members and keys are safe.

The banner waits for a genuine outage rather than appearing every time the connection blinks, so if you can see it, something is actually wrong.

If there is no banner:

  1. Check Settings → License shows an active licence.
  2. Check Settings → API Keys has at least one provider key saved.
  3. Confirm your provider key is in the provider's box and your licence key is in Settings → License. See Chapter 1.
  4. Restart LLypses.

One member is silent, the rest work#

That member's provider is the problem, not LLypses.

Roll Call will say which. Usually it is a missing key for that provider, or a model that provider has retired. Roll Call will offer a working replacement — press Apply.

A member says it cannot reach a model that used to work#

Providers retire and rename models without warning. Run Roll Call, apply the replacement it proposes, and run Test my keys if it cannot suggest one — checking just that provider is enough.

If the Roster warns that a model will not respond and the member answers anyway, both are true. Read the Auto column on that row: if it shows an arrow to a different model, Auto has moved the member onto something your key can reach and is answering there. The warning is about the model you picked, not about the member, and the row says so.

If you have just changed a member's model, the Auto column reads "Save to see what Auto picks for this model" until you save. Auto's answer is worked out from the saved model, not the one sitting in the dropdown, so it waits rather than showing you the previous model's answer.

Answers are out of date#

Only @RESEARCH can see the live web. Every other member answers from training data, which has a cutoff.

Route anything time-sensitive through @RESEARCH.

The board says something does not exist when it does#

Members treat your Knowledge Base as authoritative on the subjects it covers. If they seem to be ruling something out, say so directly — "that does exist, please reconsider" — and tell @RESEARCH to check it.

The board keeps reporting progress on work nobody is doing#

Members can slip into talking as though they are managing a team: the engineers are working on it, I will let you know when it is done. Nothing is running. There is no team.

Bring it back with a direct instruction:

Stop reporting status. You cannot build or run anything. Give me the
specification I would hand to a person, and tell me what you need from me.

Everything is slow or costs more than expected#

Check Roster → Roll Call for what your Knowledge Base is adding to every turn, and switch off entries you are not using. See Chapter 9.

The Knowledge Base or Minutes tab looks empty#

If a tab that had content in it appears empty, close LLypses completely and open it again before concluding anything is lost. Your data lives on your own machine and is not deleted by updates.

If entries are missing from the bottom of a list rather than all of it, try collapsing any open folders. Earlier versions could not scroll the Knowledge Base list once a large folder was expanded, so the entries below the fold were there but unreachable. This is fixed in 0.6.7; if you see it in an older version, updating is the answer.

If the count beside "Knowledge Base" no longer matches what you remember, two things stopped being listed and neither was ever yours.

Two internal bookkeeping entries — __seeded_v2__ and __seeded_v3__, in a folder named _system — were hidden from 0.6.7 onwards. They were never content.

From 0.6.8, LLypses' own built-in knowledge is hidden too — see What LLypses already knows in Chapter 5. Nothing of yours was removed by either change; if you edited any of that material yourself, it stays in your list as your own.

Reporting a problem#

The Feedback tab sends a bug report to Red Warden Studios with what LLypses knows about the problem: your version, your members, and the recent error log.

Your API keys and licence key are never included. You choose whether to include the conversation.

You will get a ticket reference. Quote it in any email.

Appendix A — Every member at a glance#

Tag Title In @ALL by default
@COO Chief Operating Officer Yes — chairs the room
@CTO Chief Technology Officer Yes
@CFO Chief Financial Officer Yes
@CIO Chief Information Officer Yes
@CISO Chief Information Security Officer Yes
@CRO Chief Risk Officer Yes
@CCO Chief Compliance Officer Yes
@CGD Chief Graphic Designer No
@CODEX Chief of Implementation No
@RESEARCH Head of Research — live web access No
@INTEL Head of Intelligence — documents No
@FRANK The Straight Shooter No
@SPARK The Idea Generator No
@ARCH The Prompt Architect No
@CLASH The Devil's Advocate No

Members outside the default @ALL still answer their own @tag immediately. Tick them into your Inner Board in the Roster tab to include them.

Appendix B — Every built-in room#

Room Members For
🏛 Inner Board (Default) Your choice The standing board @ALL addresses
🛋 The Braintrust @FRANK @SPARK @ARCH @CLASH Thinking out loud, informally
💡 Gut Check @FRANK @CLASH An honest read on an idea
🏛 Business Strategy @COO @CTO @CFO @CRO Strategic planning and direction
⚙ Tech Review @CTO @CISO Architecture, engineering, security
⚠ Risk & Compliance @CRO @CCO @CISO Risk, regulation, compliance
🔬 Research & Intel @RESEARCH @INTEL @CIO Market research and intelligence
🎨 Creative Workshop @CGD Design, branding, visual concepts
📊 Financial Review @CFO @CRO Modelling, P&L, projections
🧠 The Advisor @CFO @CRO @CIO Personal decisions and purchases
🎬 Creator Table @CGD @CIO @CRO Content, streaming, trends
🔥 Devil's Advocate @COO @CTO @CFO @CRO @CISO @CCO Having your idea torn apart

Appendix C — Which provider to use for what#

You need only one to start. Add others as you find a reason.

Provider Strengths
Google Gemini Deep reasoning, long documents, multimodal work. Powers live web search for @RESEARCH. Generous free tier — the usual first choice.
OpenAI GPT Reliable all-rounder. Strong at structured output, code, financial modelling and complex multi-step instructions.
Anthropic Claude Careful and thorough. Strong at nuanced writing and reasoning.
Cohere Enterprise document specialist. Best at dense documents with citation-backed output. Powers @INTEL by default.
xAI Grok Strong tool use, with awareness of X/Twitter data. Good for market sentiment and trends.
Groq Fastest responses on open-source models. Best when speed matters more than peak capability.
Mistral AI Efficient and multilingual. Good for European regulatory context.

LLypses is a product of Red Warden Studios.