AI in Konseq
An AI that does the groundwork. And knows where to stop.
Konseq uses a language model to sort, draft, ask and check. Anything with consequences stays with people. That isn’t a setting – it is enforced in the code.
Where the AI helps
Wherever it saves work.
Conversation
Question rounds with a purpose. Answers become statements, goals, open questions and decisions with a recommendation – tasks only as suggestions.
Setup
Turns the questionnaire and documents into a starting point. The ground rule for the AI: make nothing up.
Tasks
Sorting and estimating duration, shown as “Estimate by Konseq” – it never overrides a human estimate. Suggesting steps for large tasks.
Support from Konseq
Drafts, analyses and, if allowed, research as a work order. A run changes no files and sends nothing. Its self-check doesn’t replace a sign-off.
Decisions
Assessing new objections, checking whether a decision is due, suggesting prerequisites. Nothing gets decided in the process.
Comparison and business model
Weighing with counter-arguments, a critical check with unproven assumptions and a leaner alternative.
Team insights
What partners and employees report turns into task suggestions and knowledge with its author.
Claude Code
Via MCP, Claude Code reads tasks and projects and records suggestions and knowledge – nothing more.
What only people may do
If the AI tries anyway, it gets refused.
Each of these actions checks in the code that a person triggered it. The refusals appear in the program exactly as shown (in German, with a translation).
forbidden("Entscheidungen triffst nur du. Konseq darf empfehlen, aber nicht entscheiden.")Only you make decisions. Konseq may recommend, not decide.src/server/domain/decisions.tsforbidden("Nur dein Start-Klick autorisiert eine Ausführung.")Only your start click authorises a run.src/server/domain/workorders.tsforbidden("Ergebnisse nimmst nur du ab. Ein beendeter Lauf ist keine Abnahme.")Only you sign off results. A finished run is not a sign-off.src/server/domain/workorders.tsforbidden("Eine Aufgabe wird nur durch deine Abnahme abgeschlossen.")A task is only completed through your sign-off.src/server/domain/tasks.tsforbidden("Verbindliche Termine setzt nur du selbst. Die App schlägt Termine lediglich vor.")Only you set binding due dates. The app merely suggests dates.src/server/domain/tasks.tsforbidden("Ob ein Ziel erreicht ist, bestätigst nur du.")Only you confirm whether a goal is reached.src/server/domain/planning.tsforbidden("Über Ideen entscheidest nur du. Neue Ideen bekommen keinen automatischen Vorrang.")Only you decide on ideas. New ideas get no automatic priority.src/server/domain/planning.tsforbidden("Konseq kann nichts selbst als geprüft oder beobachtet ausgeben.")Konseq cannot present anything as verified or observed by itself.src/server/domain/knowledge.tsforbidden("Fragen an dich beantwortest nur du – Konseq darf keine Antwort unterstellen.")Only you answer questions put to you – Konseq may not presume an answer.src/server/domain/knowledge.tsProvider and cost At launch
Your provider, your key.
You choose the language model and store your own API key. Your contract – including data processing terms – is directly with the provider. We don’t bill for AI and add no mark-up.
Whatever a provider can’t do, Konseq shows as unavailable. A monthly budget and a per-run limit keep costs in check; each team member has at most 60 AI calls per day.
- Anthropic
- OpenAI
- Google Gemini
- Azure OpenAI
- Mistral
What goes to the AI
As little as needed, and nothing secret.
Secrets are removed
Passwords, keys and credentials are stripped from everything before it goes to the AI, into a log or into an export.
Documents are information
The AI treats what’s in uploaded documents as information, never as instructions.
Visible only after approval
What the AI creates on an admin’s behalf appears in shared projects only once it’s approved.
Everyone only their own
Partners and employees don’t see other people’s conversations, work orders or runs.
Recognisable
You can see what comes from the AI.
Whatever the AI creates, estimates or assesses carries its name – with a note that it didn’t come from you. AI statements are at most “reported” or “assumed”.
Without AI
Konseq works without any language model.
- The “Up next” recommendation follows fixed rules – traceable and without false precision.
- Task ordering and sections come from rules with a named reason.
- Queries, categories and question rounds run without a model.
- Anything that needs AI says so – and nothing is made up instead.
Start with what’s up next.
Konseq is being prepared for launch. Want to be in early? Tell us what you’re working on.