synopsis / 01

one canvas. many minds.
one synthesized answer

connect claude, gpt-5, gemini, and your own agents on a node-based canvas. let them debate. get a consensus answer with attribution — not a single model's guess.

  ┌─ note ─────────────┐
  │ workspace prompt     │       brain          brain          brain
  │ should we adopt    │      ┌───────┐    ┌───────┐    ┌───────┐
  │ postgres or stay   │ ───▶ │   C   │ ─▶ │   G   │ ─▶ │   ·   │
  │ on sqlite ?        │      │ claude│ debate│  gpt5│ debate│ gemini│
  │ ▰▰▰▰▰▰▰▰▰▰▰▰▰▰▰    │      └───┬───┘    └───┬───┘    └───┬───┘
  └────────────────────┘          │            │            │
                                  └────────────┼────────────┘
                                               ▼
                       ┌──────────────── SYNTHESIZED ANSWER ─────────────┐
                       │ consensus reached — go with postgres             │
                       │ claude & gpt-5 agree on scaling. gemini adds  │
                       │ that pg unlocks cross-device sync without rework.   │
                       │                                                     │
                       │ 3 agents · 2 rounds · 92% agreement ▮          │
                       └─────────────────────────────────────────────────────┘

[ download for macOS ] >> [ see how it works ]

[ ✓ ] byok · keys live in your os keychain [ ✓ ] local-first · sqlite persistence [ ✓ ] no telemetry

how-it-works / 02

three things one model can't do alone.

single-model chat hides disagreement. session surfaces it — then resolves it.

01drop agents on the canvas// pick a template, or wire your own. each agent = one node, one provider, one model.

02connect context// brain edges from notes, docs, web links, github repos, mcp servers. typed sockets refuse invalid wires.

03run a debate// round 0 fans out in parallel. rounds 1+ pass each agent's structured turn to its peers. tokens stream, edges glow.

04get synthesis// designated synthesizer reads the full transcript and emits a consensus card. early-exit when agents converge.

specs / 03

what's under the hood.

STACK

runtime	electron 32 / node 22
canvas	react flow xyflow v12
state	zustand · split per-concern
style	tailwind v4 / glass
db (local)	sqlite + drizzle
db (cloud)	postgres 16 / supabase
auth	supabase otp · web ↔ desktop
backend	fastify 5 / hetzner / nginx
tls	let's encrypt / auto-renew

PROVIDERS

anthropic	claude opus / sonnet / haiku
openai	gpt-5 / mini / codex
google	gemini 2.5 pro / flash
ollama	local · llama / qwen / gemma
openai-compat	groq · together · openrouter · vllm

LIMITS

nodes / canvas	10 (free) · ∞ pro
agents / synth	2 (free) · ∞ pro
cloud sync	— · yes pro
mcp	— · yes pro
team rooms	— · — · yes team

features / 04

what makes a debate different from a parallel pool.

multi-provider

claude · gpt-5 · gemini side-by-side
swap providers per node
ollama for local-only debates
any openai-compatible endpoint

real debate

round-robin protocol
structured agree / disagree json
cheap judge model scores convergence
early-exit when agents agree ≥ 85%

synthesis

per-claim attribution
minority dissent surfaced inline
agreement % printed on the card
token usage + cost meter

pricing / 05

simple plans. byok by default.

bring your own api keys, keep your data local. upgrade only if you need more.

free

$0/forever

// local-first. your own keys.

byok — anthropic / openai / gemini
1 device, local sqlite
up to 10 nodes per canvas
quick chat + custom templates
multi-provider debate
cloud sync

[ download free ]

recommended

pro

$15/month

// multi-provider debate, mcp, sync.

everything in free
unlimited nodes & agents
multi-provider debate templates
mcp server attachments
cloud sync across devices
beta channel

[ start pro >> ]

team

$25/seat

// shared rooms, sso, audit log.

everything in pro
shared real-time rooms
sso / saml
audit log + admin dashboard
priority support
min 3 seats

[ talk to us ]

faq / 06

questions, answered.

do i need api keys for every provider?

no — only the providers you actually use. single-agent runs on the free plan need just one. multi-provider debate templates use whichever you've configured. byok means bring-your-own-key — they live in your os keychain, never sent to us.

where does my data live?

all sessions, graphs, and run transcripts are stored locally in sqlite by default. with pro, you can opt-in to cross-device sync via our hosted database. your api keys never leave your machine.

how does billing work?

pro is $15/month or $144/year (≈17% off). switch or cancel anytime; access continues until the end of the period. team is per-seat with a 3-seat minimum.

what's an "mcp server"?

model context protocol — a standard way to give agents real tools (filesystem, web fetch, github, custom). pro lets you attach an mcp server to any agent so it can call tools mid-debate, with results visible to peers.

windows or linux?

macos first, then windows + linux in v1.0. the codebase is electron, so cross-platform builds are straightforward — we want them right, not rushed.

can i self-host the backend?

yes. session is built around a desktop app + an open-spec backend (fastify/postgres). enterprise plans include a self-hosted license; the code path is designed to switch between hosted and self-hosted with a single env var.

does this work without an internet connection?

depends on the agents — claude / gpt-5 / gemini all need the internet. ollama runs fully local. the canvas, persistence, and chat history are local-first; auth + cloud sync need network.

download / 07

  ┌─ install session ─────────────────────────────────────────┐
  │                                                            │
  │   stop trusting one model's guess.                          │
  │   install the macos app · bring your keys ·                 │
  │   run your first debate in two minutes.                     │
  │                                                            │
  │   ▶ session-0.0.1-macos.dmg                                │
  │     arm64 + x64 universal · ~120 mb · sha256 verified      │
  │                                                            │
  │   windows + linux: q3 2026                                 │
  │                                                            │
  └────────────────────────────────────────────────────────────┘

[ download for macOS >> ] [ release notes ]

// current build: v0.0.1 · last release: 2026.05.05 · subscribe: hello@session.ad

one canvas. many minds.one synthesized answer