by Synenergy · cases

Which mode pays off — and which does not.

“Your documents” is for large context with quotes and a visible cost — mainstream recursive reading of files you supply. The panel is for a plural second opinion. Pick by the job.

← Modes on homeJoin the waitlist

At a glance

ModeBest forPanel and cost
Your documents (Gather & Judge)big context, contradictions, cost-visible runsone run, cost shown
Comparequick sanity check — do the models agree?several models, no judge
Deliberatedaily work — one reliable answer instead of several raw onespanel plus one judge
Votevalidating a finished decision, text, or code before releaseeach model votes alone
Brainstormfinding new ideas when you need a fan of optionsblind generation, then recombine
Extendeddecisions where shared blind spots are dangeroustwo independent channels
Councilarchitecture, strategy, naming — hard reversalsone structured session
Route Laboperators and agent builders setting lasting defaultsfree scouting; paid runs need approve

Illustrations are schematic, not product output. The parameters they show (accept / reject / abstain, two independent channels, free versus paid Route Lab, cost shown per run) are those listed on the Consilium home page.

How to read this

Each case is one mode: the payoff, a concrete situation, and an anti-case. Broader is not always more accurate. A quote-check proves the quote is in your files — not that the conclusion is right.

Cases by mode

01

Your documents (Gather & Judge)

Illustration: How Your documents worksSeveral files are worked through in steps, ending in an answer with a quote from the files and a cost meter.“filesstepsquotecost shown
Illustration
Payoff
work a large file set without dumping it into one expensive chat — quotes, a quote-check, and a clear cost. Cheap helpers prep; judgment stays controlled.
Situation
specs, notes, and logs that will not fit one prompt; you need “what do these files actually say about X” with quotes.

Best for: big context, contradictions, cost-visible runs.

Not for: second-opinion panels, one short file you can read yourself, or world facts with no files. A passed quote-check means the quote is in your files — not that the conclusion is true worldwide.

02

Compare

Illustration: How Compare worksThree models answer the same prompt independently and the answers sit side by side, with no judge merging them.ABCside by side, no judge
Illustration
Payoff
see disagreement before you commit — who hallucinates, who hedges.
Situation
you have an open design or strategy question and want side-by-side answers without a judge collapsing them yet.

Best for: quick sanity check — do the models agree?

Not for: a single fused answer (use Deliberate).

03

Deliberate

Illustration: How Deliberate worksSeveral answers are distilled by a judge into one answer, with the points of disagreement marked rather than hidden.answersone answer◆ disagreement stays visible
Illustration
Payoff
one distilled answer plus visible contradictions — everyday second opinion.
Situation
you need a usable write-up after the panel, not a raw matrix of voices.

Best for: daily work — one reliable answer instead of several raw ones.

Not for: unverified world facts; not a substitute for reading your own files (use Your documents).

04

Vote

Illustration: How Vote worksA finished candidate goes to three models; each returns its own accept, reject or abstain, shown as a tally with no single merged verdict.candidatetallyno fused verdict
Illustration
Payoff
an independent accept / reject / abstain gate on a finished candidate — no fused essay.
Situation
a PR, release note, or decision text is ready; you want a quality signal before merge.

Best for: validating a finished decision, text, or code before release.

Not for: replacing tests or human review; not for open ideation.

05

Brainstorm

Illustration: How Brainstorm worksModels generate ideas without seeing each other, the strongest are recombined, and every claim stays marked unverified.blind ideasrecombineunverified
Illustration
Payoff
a fan of options, not one opinion — blind generation then recombination.
Situation
naming, positioning, or product directions where you need divergence first.

Best for: finding new ideas when you need a fan of options.

Not for: checking contradictions across a set of files (use Your documents).

06

Extended — West vs East

Illustration: How Extended worksModels from a Western and an Eastern school reach one panel through two independent channels.two channelsWestEastone panel
Illustration
Payoff
cross-school coverage when one lineage’s blind spots are expensive.
Situation
a high-value open question where shared training priors could hide the same miss.

Best for: decisions where shared blind spots are dangerous.

Not for: cheap chat; not a statistical independence guarantee.

07

Council

Illustration: How Council worksA panel meets in one structured session and ends in a single decision.panel sessiondecision
Illustration
Payoff
a structured panel session for calls that are expensive to reverse.
Situation
architecture, strategy, or naming at a fork you will live with for months.

Best for: architecture, strategy, naming — hard reversals.

Not for: routine edits or single-file questions.

08

Route Lab — evidence before you hard-code a default

Illustration: How Route Lab worksA task class is scouted for free, then candidate routes are run in a small paid experiment you approve and ranked.free scout◆ paid run, approveranked routes
Illustration
Payoff
ranked evidence for which solo model or panel recipe fits a class of tasks — before you ship that choice into Hermes, Cursor, or another agent.
Situation
you’re wiring an agent and don’t want to guess the default model or route from a blog post. Free Lab helpers scout the catalog; a tiny paid experiment ranks real candidate routes (approve spend).

Best for: operators and agent builders setting lasting defaults.

Not for: one-shot user questions; live A/B on every chat; accept/reject of a finished artifact (use Vote). Runtime answers still use compare, deliberate, vote, or Your documents.

How agents pick a mode

Connect once. Prefer “Your documents” when the pile will not fit one prompt (quotes + cost control). Use the panel for a plural second opinion. Use Route Lab when choosing a lasting default for a task class — not for one-shot chats. After an update, reconnect once so new tools appear. A normal single-model chat link alone does not run “Your documents” or Lab.

Connect / verifyAgent scenariosJoin the waitlist