One AI on its own is rarely up to an important decision. The strength appears once several AIs argue it out properly: contradictions surface, and a consensus that holds is reached.
The Plenum skill earns its keep when a matter is complex and a single AI would decide too quickly — or on a poor factual basis. In the Plenum several AI models discuss the matter and expose each other's strengths and weaknesses; five as shipped, up to eight are possible. The built-in cross-examination then presses much harder, which brings out problems that are otherwise overlooked or quietly passed over. Trouble is nipped in the bud, and programming work is thought through properly from the start. On the principle: save ten per cent on preparation and you pay eighty per cent in clean-up.
New: a bias filter. It keeps arguments out of the assessment that do not rest cleanly on facts — and that would otherwise push an important decision the wrong way from the very beginning.
All prices exclude VAT; 19 % German VAT is added. Sales are exclusively to business customers.
In independent product review
SYN/PSV1.1.0
ABCΣ
Choose licenceV1.1.0
One-time payment · 24 months of updatesAll prices exclude VAT; 19 % German VAT is added. Sales are exclusively to business customers.
Coming soon
The product is finished and reviewed. We will open sales once payment handling and fulfilment have been fully signed off. Until then you can see the full scope, the price, and every condition here — without a purchase option.
Work product
What changes after using it
Noticeably fewer bad decisions, because the matter is thought through early and thoroughly — and because up to eight AIs have to argue against each other instead of agreeing with you.
+Included
✓Five rounds of cross-examination: you do not see five opinions, you see what survives the mutual attack
✓Bias filter: arguments without a clean factual basis are thrown out before they distort the decision
✓A check on your input before the first call costs money: the AIs tell you in writing whether your basis is even sufficient to decide
✓Fixed spend ceiling: the run stops before it costs more than you allowed
✓Runs on your own machine. Your text goes only to the models you pick yourself
✓AI.md: your own AI reads the manual itself and sets everything up for you
✓English PDF manual with complete examples in German and English
✓Run interrupted? It picks up where it stopped and never charges you twice
✓One skill for Claude Code and Codex. No new program to learn
–Deliberately not included
×OpenRouter access, model credit, or API charges
×Legal, tax, medical, or investment advice
×Automatic approvals, contracts, or system actions
Workflow
From input to a reviewable result
01
Frame the decision
Capture the goal, options, criteria, and red lines in one concise brief.
→
02
Assess independently
Several model families evaluate the same case separately within the spend ceiling.
→
03
Synthesize dissent
The chair exposes consensus, minority positions, uncertainty, and the next checks.
→
From practice
Customers who have worked with it
At first I thought it was just another chat window with a better name. It is something else: I write down what I want to decide, and at the end I get a letter telling me what speaks against it. That was the difference for me.
Stefan MerzHolzbauTranslated from German
I had already made up my mind and really only wanted confirmation. Three of the models followed my direction, one did not — it insisted on a point I had considered marginal. I brushed it aside at first and came back to it that evening. It was right, and I am glad I had it in writing.
Evelyn BiesingerMöbeldesignTranslated from German
As a managing director you rarely get genuine pushback, to be honest about it. Here I had to defend my preferred option against objections I had not thought of. In the end I still decided as planned — but afterwards I knew why.
Alexander RotterManaging Director, RohrbauOne-to-one coaching client of SynclaroTranslated from German
Setting it up cost me an evening, and you need your own account with the model provider — you should know that beforehand. After that it runs locally for me, I see what each run is allowed to cost before it starts, and for this kind of question I now use it regularly.
Daniela BeischHolzbau Ott, GammertingenTranslated from German
Every statement was released by the named person. They describe personal experience and are not a promise of any particular outcome.
Requirements
Python 3.10 or newer
Your own funded OpenRouter account with a key that has a spend limit set; without that limit the runner refuses every paid call
Models chosen from providers offering endpoints without data retention — verified 16.08.2026: Anthropic, Google, xAI and DeepSeek. OpenAI and Qwen models cannot be used
Willingness to send brief content through OpenRouter to your selected model providers, possibly outside the EU
macOS, Linux, or Windows with WSL2 and basic familiarity with JSON files
Tested compatibility
CodexClaude CodemacOSLinuxWindows with WSL2
✓A good fit for
Make-or-buy and architecture decisions
Vendor, tool, and strategy selection
Teams that want dissent before approval
×Not a fit for
Simple fact lookups with one authoritative source
Fully automated decisions without human accountability
Personal or highly sensitive data without your own privacy review and suitable model-provider terms
Disclosure
Data flow and running costs
Data flow during operation
OpenRouter is an intermediary: a single point of access through which you reach models from Anthropic, Google, xAI, DeepSeek and others without holding an account with each provider. During operation, your decision brief and the role briefs travel that way to the providers you select. Depending on your model choice, processing may also take place outside the EU. Some providers offer endpoints that verifiably do not store your content — zero data retention. The runner demands that mode on every single call and additionally denies data collection. If no suitable endpoint exists, it aborts rather than quietly falling back to one that stores. That is the built-in safety line, and it is also why not every model listed on OpenRouter can be used. What that safety line does not do, we say just as plainly: it does not make your content anonymous, it does not replace a data-processing agreement, and it does not guarantee EU-only processing. Details and limits are described in the bundled "Privacy and security" documentation.
Running model costs
Every plenum run incurs model charges on your own OpenRouter account. As a non-binding guide without warranty: roughly EUR 1 to 5 per run. Cheaper models put a run near the lower end, frontier models near the upper. Actual charges depend on model choice and the number of rounds and may change with provider pricing. The built-in spend ceiling caps expenditure before every single call.
Clear before purchase
Operations, support, and delivery
01
Support
Deliberately lean: the bundled AI.md briefs your own AI, and the manual answers the rest. There is an email address for reproducible product defects. Personal application consulting is not included; the Team and Company tiers add one setup question and one briefing session respectively.
02
Updates
The purchased version remains usable indefinitely. Updates that preserve functionality — security fixes and adaptations to changed OpenRouter endpoints and model identifiers — are provided for 24 months from delivery. You agree to this limitation explicitly at purchase. You can change model names and endpoints yourself in the configuration file at any time; the manual explains how.
03
Delivery
Buyer-marked ZIP, licence record, and checksum in your private account — immediately after payment.
Licence model
Use it. Do not resell it.
The licence permits use within the purchased scope. Source files, skills, software, and templates may not be published, sublicensed, or passed on as your own product.
Some things only make sense once you can see them. Describe the structure in one sentence and get it back as a hand-drawn sketch you can drop into a proposal, a concept or a training deck.