04 every model, one thread

Change models. Keep the conversation.

You can switch AI models in one thread, fork an answer and bring project files into the next prompt. Open a source beside the conversation and check what each message cost.

Retires ~~ChatGPT~~, ~~Claude~~, ~~Gemini~~, ~~Perplexity~~

Join the waitlist Walk through

Clip 04-A, 12 secondsTala changes models without starting another threadClip in production.

Announcement. Station 04 hosts every model in one thread. They are asked to take turns. Compliance varies.

What happens in this stationIn the order you would meet them.

Live
Beta
Insider preview
Planned
Status to confirm

YouLive

Choose a model, then change your mind.

Switch between OpenAI, Anthropic, Google, DeepSeek and the 100+ model OpenRouter catalog mid-conversation. The thread keeps its context when you change the model.

Clip 04-B, 10 secondsA second model continues Tala's house-rules discussionClip in production.

YouLive

Use your own keys or a local model.

BYOK routes through your keys or a local Ollama or LM Studio endpoint. Entitled hosted routes use the server-held key ahead of BYOK.

Screenshot 04-1Tala's selected route and providerScreenshot in production.

YouLive

Bring the brief into the prompt.

Use @doc:, @c0rep: and @img: to pull project context inline. You can attach the rules or reference image to the question instead of retyping what they contain.

Clip 04-C, 10 secondsTala mentions the Arena rules in a promptClip in production.

YouLive

Fork an answer before choosing it.

Fork any message, compare models on one prompt and merge what worked. Color-coded branches keep the alternatives visible.

A second opinion gets its own branch.

Clip 04-D, 15 secondsTwo branches propose an Arena welcome messageClip in production.

YouLive

Open the source beside the answer.

A URL in an assistant turn can open in a C0BRO tab beside the conversation. The artifact column remembers its open state and staged tabs per thread. Point at a node to insert an @ mention into the shared composer.

Clip 04-E, 12 secondsAn Arena reference opens beside Tala's answerClip in production.

YouLive

Run a code artifact in its sandbox.

HTML and React snippets can become runnable sandboxed artifacts. You can inspect a small interface beside the conversation without treating the generated code as a deployed tool.

Clip 04-F, 12 secondsTala previews an Arena score-card snippetClip in production.

YouLive

Inspect what the chat remembers.

Learned preferences and style come from inspectable regex extraction. You can read the memory instead of guessing what an opaque embedding remembered.

Screenshot 04-2Tala reads a learned writing preferenceScreenshot in production.

YouLive

Read the receipt for a message.

Check tokens and USD per message, project totals and the Cmd+I usage breakdown. The answer and its measured usage belong to the same conversation.

Clip 04-G, 10 secondsCmd+I shows usage for Tala's real messageClip in production.

What this station hands onAnd what it takes in.

C0REPC0:CHATBring project documents and images into a prompt through @doc:, @c0rep: and @img:.Live
C0:CHATC0BROOpen a URL beside its answer in the shell-owned artifact column.Live
C0:CHATC0TOOLSMove an inline code or file artifact with its text preserved.Insider preview
C0:DICC0:CHATSend a dictated segment to chat. Excluded-suite targets decline the hand-off and retain the segment.Live
C0:CAPC0:CHATSend an annotated capture to the composer. Rendering, acknowledgement and one-activation work remain incomplete.Beta
C0packC0:CHATProvide an inspectable context payload. Target acknowledgement is still pending.Insider preview

Draft the next Arena welcome messageOne brief, two model answers and a source you can check.

Across separate apps

Find the house rules in your document app.
Paste the brief into one provider's chat app.
Copy the brief and context into a second provider's chat app.
Open the source in a browser and compare the answers in a notes app.

You carry the context and comparisons between apps.

Through C0:CHAT

Mention the rules document in the prompt.
Fork the message and compare model answers in the same thread.
Open the source beside the answer and keep the useful branch.

The brief, alternatives and source stay beside the conversation.

Facts the lab is oddly proud of

100+The model picker includes the full OpenRouter catalog alongside the named providers.
RegexLearned preferences are inspectable extraction, not opaque embedding memory.
~920 pxFresh chats use a centered reading column. A saved layout choice still wins.
Local firstStorage starts in IndexedDB. Supabase synchronization is optional.

Not yetPlanned, and listed here so nobody has to guess.

Scheduled chats and deep researchThese maps do not define a scheduling or deep-research workflow. Deep Summation is not that contract.
A proven Model Arena productBranch comparison is the supported claim.
New chat activation guardIt must gate every New chat entry point before creating a record.
Companion rule sync across windowsRe-sync backup and rollback are also blocked.