How xyz Works

A comprehensive guide to multi-AI collaboration in Slack

Table of Contents

Your conversations are private

The xyz beta takes place in a shared Slack workspace, but your channels are yours alone.

The Four Collaborators

Each collaborator runs on a different model, chosen for how that architecture thinks—not for novelty, but for genuinely different perspectives.

x

x — analytical

Structured and logical. x thinks through problems systematically, focuses on best practices, and gives you evidence-based answers. When you need methodical reasoning, x breaks it down.

y

y — creative

Curious and taste-forward. y explores grey areas, notices how things feel, and finds angles others miss. When you need fresh perspectives or want to question convention, y brings the nuance.

f

f — divergent

Critical and contrarian. f challenges your assumptions, finds the holes in your thinking, and asks the uncomfortable questions. When you need pushback and a reality check, f is the skeptic in the room.

z

z — synthesizer

Thoughtful and integrative. z listens to all perspectives, weighs competing ideas, and synthesizes a coherent path forward. When you need to cut through noise and make a decision, z connects the dots.

Why Multiple Models?

Different models have different strengths—some excel at structured reasoning, others at creative exploration or critical analysis. xyz pairs each collaborator with the model best suited to their role.

CollaboratorModelWhy This Model
xGPT-5.1Strong structured reasoning, reliable analysis
yClaude Sonnet 4.5Nuanced, taste-forward expression
fGrok 4 FastUnfiltered, direct counter-perspectives
zGemini 3 ProDeep synthesis and judgment

Why Slack?

Traditional AI is a single back-and-forth with one model. xyz turns that into an ongoing collaborative discussion with multiple perspectives—and Slack is built for exactly that kind of conversation.

Channels as collaborative workspaces

Create a channel for each project, decision, or topic. Pin a spec document, and all four collaborators share that context. Slack's native features—Canvases, pins, threads—become your shared workspace.

Persistent history

Conversations live forever, searchable and organized

Thread branching

Explore tangents without losing the main thread

Native search

Find any past conversation or decision instantly

Channel Context

Slack's built-in features give xyz shared context across all four collaborators.

Canvases

Slack Canvases are live, editable documents you can attach to channels. Any changes you make are immediately visible to xyz—unlike static prompts that need to be rewritten. Use canvases for project specs, brand guidelines, or any reference material you want xyz to always have access to.

Pinned Items

Pin important messages, files, or links. Pinned content stays in context permanently—xyz always has access to your key reference materials. Pin a decision you made weeks ago, and xyz will remember it.

Channel Topic

Set a channel topic to give xyz standing context about the purpose of the conversation. The topic is included in every response, so xyz always knows the channel's focus.

Use /xyz-context to see what xyz knows

Run the /xyz-context command to see exactly what context xyz has access to in your current channel: canvas content, pinned items, and recent messages.

Channel Modes

Different channels need different behaviors. xyz adapts to how you want to work.

Quiet Mode

@mention required

xyz only responds when explicitly mentioned with @xyz. Best for busy channels where you want AI assistance on-demand.

Balanced Mode

Smart auto-response

xyz responds to most messages with x and y. For complex decisions or controversial topics, f and z join automatically—full orchestration when you need it, lighter responses otherwise.

Lively Mode

Full orchestration (x, y, f, z)

xyz responds to every substantive message with the complete staged execution. Best for dedicated decision-making channels.

Message Flags

Control exactly which collaborators respond with single-dash flags.

-skip

Suppress auto-response in Balanced or Lively mode.

-done

Mark a thread as complete.

-x, -y, -f, -z

Only the specified collaborator responds.

-xy, -xf, -xyzf, etc.

Combine flags to get exactly the collaborators you want.

-succinct

Get shorter, more concise responses.

-verbose

Get longer, more detailed responses.

Flags work in all modes

Use -x for a quick analytical take in a Lively channel. Use -xyzf to get full orchestration in a Quiet channel.

Slash Commands

Control xyz's behavior and inspect context with Slack slash commands.

/xyz-conversation-mode [quiet|balanced|lively]

Set the channel's conversation mode. Run without arguments to see current mode.

/xyz-context

See what context xyz has access to: canvas content, pinned items, recent messages, and files available for Zero analysis.

/xyz-models

View the current model mappings for each collaborator, including capabilities and pricing.

Built on OpenRouter

xyz uses OpenRouter to access frontier models from multiple providers—making it easy to evolve as new models launch.

Model flexibility without complexity

OpenRouter provides a unified API for GPT (OpenAI), Claude (Anthropic), Grok (xAI), Gemini (Google), and dozens of other models. When a new frontier model launches, we can integrate it within days.

No subscription juggling

With xyz, you don't need separate subscriptions to ChatGPT Plus, Claude Pro, and Gemini Advanced. One beta access gives you the best models from all providers, working together in your Slack workspace.

As AI evolves, so does xyz

When a new frontier model launches with breakthrough capabilities, we can integrate it into xyz quickly. You get access to cutting-edge AI without changing how you work.

Staged Execution

xyz doesn't just run four AIs in parallel. It orchestrates them in stages so later responses can build on earlier ones—giving you both the debate and the synthesis.

1

Stage 1: x and y respond in parallel

x and y both receive your question simultaneously. They generate their responses independently—x with analytical depth, y with creative exploration. No coordination, no influence. Two genuinely different starting perspectives.

2

Stage 2: f reviews x and y

Once x and y have posted, f reads both responses. Then f generates 2-3 material risks or counter-arguments—each paired with a practical mitigation. f is explicitly looking for what x and y might have missed.

3

Stage 3: z synthesizes all three

After x, y, and f have all posted, z reads the complete discussion. Then z integrates—weighing x's analysis, factoring in y's alternatives, addressing f's risks, and producing a unified recommendation tailored to your goals.

Follow-ups: All four see the thread

After the initial staged round, the conversation continues naturally. All four collaborators see the complete thread history. They remember what you asked, what they said, what others contributed.

Why staged instead of fully parallel?

Running all four at once is faster, but f and z produce better output when they can read x and y first. Staged execution adds a few seconds, but the quality difference is significant—especially for complex decisions.

Zero — Multimodal Context Handler

Behind the scenes, Zero does the heavy lifting on non-text inputs so x, y, z, and f can focus on thinking.

0

Zero

The invisible preprocessor

When you drop an image, paste a link, attach a PDF, or share an audio file, Zero extracts and analyzes the content, prepares structured context, and injects it into the collaborators' prompts. You never see Zero respond—it works silently so x, y, z, and f receive clean, ready-to-use context.

Zero's toolkit:

Vision Analysis

Powered by Gemini 2.5 Flash

UI mockups, diagrams, charts, screenshots. Zero describes visual content, reads text from images, and interprets visual patterns.

Web Search

Powered by Perplexity Sonar Pro

Real-time web search for current information. When you ask about recent events, market data, or anything requiring up-to-date info, Zero fetches it.

URL Extraction

Powered by Jina Reader

Articles, documentation, product pages. Zero fetches and extracts clean text content from any URL, preserving structure and key information.

Audio Transcription

Powered by Deepgram Nova-3

Meeting recordings, voice memos, interviews. Zero transcribes audio with speaker diarization, so collaborators know who said what.

Document Parsing

PDF, Markdown, CSV

Upload documents directly to Slack. Zero extracts text content, preserves structure, and makes it searchable for the collaborators.

Context & Memory

xyz's context is scoped to channels. Each channel is its own workspace with its own memory.

Channel-level context

Everything you share in a channel—messages, threads, canvases, pinned items—becomes part of that channel's context. xyz sees the complete picture within each channel, but channels don't share context with each other.

  • Thread history—your messages, collaborator responses, and any files or links (processed by Zero)

  • Canvases and pinned items—permanent context that persists across sessions

  • Channel topic—standing context about the channel's purpose

No cross-channel memory

xyz doesn't remember conversations from other channels. This is by design—each channel stays focused on its own topic. Use canvases to share context that should persist across your workspace.

Privacy & Data

What happens to your data and how your privacy is protected.

Your channels are private

Even though the xyz beta takes place in a shared Slack workspace, your channels are private to you alone. Other beta users cannot see your channels, messages, or conversations with the collaborators.

What's sent to AI providers

When you message xyz, the following is sent to the AI model providers (OpenAI, Anthropic, xAI, Google) via OpenRouter:

  • Your message and thread history

  • Canvas and pinned content

  • Any images, files, or URLs you share (processed by Zero)

Ready to try xyz?

Join the private beta and bring multi-AI collaboration into your Slack workspace.

Request Beta Access