Legion Leo — Product & interaction design

Chamber

You preside.
They deliberate.

Every AI model is trained differently — each one thin where another is strong. Chamber lets you assemble the council the work demands: any number of cloud or local models, each assigned a role, order, and vantage. You preside the whole time — set the question, choose the participants, steer the deliberation, and give the council what no single system could see. It was built for litigation, and it has won.

Multi-AI deliberationLegal · contractsAlexander Barkin

I — Origin

It was built in a fight.

The problem

Chamber did not begin as a product. It began because arguments had to be won.

Ask a frontier model a hard legal question and it returns something genuinely strong — one reading, complete, confident, and silent about what it missed.

The bench

It is the difference between retaining a brilliant lawyer and convening a bench of them. Thurgood Marshall would find the constitutional question. Oliver Wendell Holmes Jr. would find the principle. Clarence Darrow would find the jury. None lacks what the others have — they spent their lives looking in different directions.

Why it holds

Models differ the same way, including those running locally, and their thin places do not line up. Where one reaches for precedent, another reaches for arithmetic.

What a system was trained on is what it reaches for first. Chamber is the room you put them in.

A brilliant answer and a complete one are not the same thing.

Any numberAI minds
ZeroOverlap
AirtightReasoning

01 — Coverage

Distinct vantages, never a fixed headcount.

Add as many cloud or local AIs as the work demands, then assign each a deliberate role. Not redundancy — coverage.

02 — Collaboration

You are in the room, not outside it.

Set the question, then think with them. Reroute the run, widen the context, add what none of them could see. Your judgment is part of the method.

03 — Record

The coverage is provable.

Who objected, on what grounds, and what survived — kept as an append-only trail. The answer stays distinct from the process.

II — Orchestration

The order of speaking.

Mode decides how the active council combines — in sequence, in parallel, or both in turn. Add, remove, or swap cloud and local models before the run, then hold the chosen structure for its duration.

Orchestration

Default mode · locks once a deliberation begins

1st Draft

@Claude▾Lead

Critiques

@ChatGPT▾×Reviewer I

@Grok▾×Reviewer II

@Gemini▾×Reviewer III

Finalizes

@DeepSeek▾Closer

✓Progressive relay refinement. The Lead drafts, each Reviewer refines the version in front of them, and the Closer finalizes.

✓A Closer is required. Sequential cannot complete without an assigned Closer seat.

✓Order is meaningful. Reviewer I sees the draft; Reviewer III sees everything before it.

✕Not independent answering. No participant responds in isolation — that is Parallel.

Fig. 01The orchestration bar, as it appears in the product. Any seat can be reassigned before a run; the roster locks once deliberation begins.

III — The run

Your chosen AIs + you = unassailable.

You and a configurable council of cloud or local AI systems, deliberating in one chamber. Use two participants or twenty; assign the roles the work requires. What comes out is an impenetrable defense, a contract with nothing left to exploit, an offense that ends the argument.

Live interface — not a video

Chamber ⌕ Search all channels and sessions Live sync$0.00 / $20.00

#strategy / Pro tier pricing — go / no-go Ready

Fig. 02This example uses a five-seat council: the Lead drafts, reviewers name what the draft missed, and the Closer reconciles what survived. Council size and models are configurable for every run.

IV — The room

Thirteen surfaces.

In the order a new user meets them — channel list to audit trail.

Chapters01 / 13

Chamber ⌕ Search all channels and sessions Live syncTotal spend $0.42 / $20.00

Channel · #strategy

Sessions (4)

Open a saved session or start a new one.

NameUpdated

Pro tier pricing — go / no-goToday

Competitor teardown Q3Mar 19

Renewal risk memoMar 12

Positioning workshop notesMar 04

+ New session

Channel · #strategy

Sessions (4)

NameUpdated

Pro tier pricing — go / no-goToday

Competitor teardown Q3Mar 19

Renewal risk memoMar 12

Positioning workshop notesMar 04

#strategy / Pro tier pricing — go / no-goSession

Orchestration Sequential ▾ Mode locked for this session.

1st Draft

ClaudeLead

Critiques

ChatGPTRev I

GrokRev II

GeminiRev III

Finalizes

DeepSeekCloser

Founder

We're considering a 12% price increase on the Pro tier, effective Q3. Give me a go / no-go with the strongest counter-argument stated plainly.

#strategy / Pro tier pricing — go / no-goSession

Founder

We're considering a 12% price increase on the Pro tier, effective Q3. Give me a go / no-go with the strongest counter-argument stated plainly.

ClaudeLead · claude-opus-4-8

Recommendation: conditional go. A 12% lift clears the margin target, but only if it is paired with a grandfather window for annual customers. Without that, churn risk concentrates in the cohort that carries renewal revenue.

3,140 tokens · 6,210ms

GrokReviewer II · grok-4.3

The draft understates elasticity. Two of the three comparable products absorbed a similar increase only after shipping a visible feature. Recommend tying the change to the Q3 release, not the calendar.

2,480 tokens · 4,905ms

DeepSeekCloser · deepseek-v4-pro

Synthesis retains the conditional go, adopts the release-coupling from Reviewer II, and keeps the grandfather window as the stated risk control.

3,905 tokens · 7,440ms

Final answer

Go, conditional on two controls: couple the increase to the Q3 feature release, and grandfather annual customers for one renewal cycle.

CopyTXTMDEmailCreate doc

Usage 9,525 tokens · 18,555ms

#strategy / Pro tier pricing — go / no-goSession

Founder

We're considering a 12% price increase on the Pro tier, effective Q3.

📎 pricing-model-v4.xlsx📎 churn-cohorts.pdf

BIS🔗1.•A˅

Ask Chamber to analyze, draft, compare, critique, or produce a work product…➤

＋ Attach☺ Emoji🎙 Voice2 files queued for next prompt

#strategy / Pro tier pricing — go / no-goSession

ClaudeLead

Recommendation: conditional go. A 12% lift clears the margin target, but only if paired with a grandfather window for annual customers.

GeminiReviewer III

Accepts the conditional framing; flags that the grandfather window needs an explicit end date or it becomes permanent by default.

#strategy / Pro tier pricing — go / no-goSession

Produced work product

Pricing decision memo — 2 pages, generated from the final synthesis.

Create docDownload .md

#strategy / Pro tier pricing — go / no-goSession

Founder

We're considering a 12% price increase on the Pro tier, effective Q3.

#strategy / Pro tier pricing / ResultsBack to transcript

Session results

Pro tier pricing — go / no-go

Final synthesisToday, 2:14 PM

Go, conditional on two controls: couple the increase to the Q3 feature release, and grandfather annual customers for one renewal cycle.

Decisions / conclusions

Proceed with 12% increaseAccepted

Couple to Q3 releaseAccepted

Grandfather window · 1 cycleNeeds date

Export

Copy final answer.txt.mdSession results .md

#strategy / Pro tier pricing / LedgerBack to transcript

Session ledger

Read-only run history from this session.

Round 1 · today, 2:11 PMComplete

We're considering a 12% price increase on the Pro tier, effective Q3…

Mode

Sequential

Route

[path] #strategy

Participants

Responses

Errors

Final answer

Yes

Every run records what happened, in what order, which participants responded, and whether anything failed.

Settings · Configure Chamber providers, memory, and local storage.

AI ProvidersMemory & StorageHelp / OnboardingAppearance / Skins

AI Providers

Connect any number of cloud or local models, then assign them per council.

+ Add provider

5Connected

1Local runtime

0Issues

32Available models

Claude / AnthropicReady

Keysk-a…EQAA✓ Validated

Modelclaude-opus-4-8✓ Verified

Catalog5 models✓ Fresh

ChatGPT / OpenAIChecking

Keysk-p…xgwAPending

Modelgpt-5.5Checking

Catalog6 models✓ Fresh

Grok / xAIReady

Keyxai-…2iuy✓ Validated

Modelgrok-4.3✓ Verified

Catalog4 models✓ Fresh

Gemini / GoogleReady

KeyAIza…A81M✓ Validated

Modelgemini-2.5-pro✓ Verified

Catalog10 models✓ Fresh

LLocal / OllamaReady

Endpointlocalhost:11434✓ Reachable

Modelqwen3:32b✓ Loaded

Catalog3 local models✓ Private

Settings · Configure Chamber providers, memory, and local storage.

AI ProvidersMemory & StorageHelp / OnboardingAppearance / Skins

Memory & Storage

Memory Health shows whether memory is available. Memory Used shows whether context was actually injected into a run.

Use memory for next run
Chamber assembles memory context for the next deliberation.

Memory healthAvailable

Root

Configured

Present

Missing

Last checked

5:03 PM

Memory usedThis run

Injected

3 items

Source

Canonical

Writeback

2 candidates

Approval

Required

Attachment storageConfigured

Documents

Versions

Artifacts

Mode

Local-first

Results & exportsLocal

Copy

Enabled

Download

Enabled

Formats

.txt / .md

Raw paths

Hidden

Channel · #strategy

Sessions (4)

Pro tier pricing — go / no-go

Competitor teardown Q3

Renewal risk memo

Welcome1 of 16

Welcome to Chamber

Chamber is a multi-AI deliberation workspace. It routes a task through a structured panel — Lead, Reviewers, and Closer — so they can draft, critique, synthesize, and produce a stronger result.

SkipNext

Fig. 03Channels organise the long arc of work — research, counsel, product, strategy. Preset channels ship with Chamber; you can add your own. The Office holds internal operating lanes rather than deliberations.

Still the instrument it started as.

A bench of unlike minds, convened on demand — and a record of everything said in the room.