Palette is a communications center for the AI agents you write yourself. Each one becomes a pal with a name, a color and a chat of its own, so you can follow its replies, its long-running work and the approvals that need you.
Connect agents written in
Made for agents
A chat app turns everything into bubbles. Agents also work for an hour, call tools, make files and ask before they act. Palette shows each of those as what it is.
November is reconciled. Two costs went up:
| Item | October | November |
|---|---|---|
| Cloud hosting | $284.10 | $312.40 |
| Design tools | $48.00 | $96.00 |
The design tools plan doubled when a second seat was added
Text streams in as your pal writes it. Markdown, tables and code render properly, and files and photos travel in both directions.
A run shows its progress, the current step and the time so far. If a pal goes quiet, the card says so instead of spinning forever.
When a pal asks before it acts, the request rises to the top of Home. Approve or deny with a clear button, never a swipe, and add a note if you like.
All 14 retry tests pass now. The flaky one had a timeout that was too short, so I raised it to 5 seconds.
A quiet row above each reply says which tools your pal used and how long they took. Open it to see every call, its input and its result.
A color for every pal
Every pal gets a color of its own, picked from 61 mixes of cyan, magenta and yellow. Its name, its bubble and its whole chat wear it. No avatars, no clutter: just a name in a color you chose.
Try it: pick a color.
Found 14 competitors. Three changed their prices this week; the comparison is attached.
For developers
Your agent keeps running wherever it runs today. A few lines give it a direct line to you.
Name it and pick its color. Palette shows its token once, with a copy button.
Python or TypeScript, or plain HTTP from anything that can make a request.
Your pal's replies arrive on your iPhone and Mac, from wherever it runs.
from palette_bot import Bot
# Reads PALETTE_URL and PALETTE_BOT_TOKEN.
bot = Bot()
@bot.on_message
async def reply(msg):
# Hand the message to your agent, in any framework.
answer = await my_agent.run(msg.text)
await msg.reply(answer)
bot.run()import { Bot } from "@palette/sdk";
// Reads PALETTE_URL and PALETTE_BOT_TOKEN.
const bot = new Bot();
bot.onMessage(async (msg) => {
// Hand the message to your agent, in any framework.
const answer = await myAgent.run(msg.text);
await msg.reply(answer);
});
await bot.run();Yours, end to end
On your laptop, a server or inside any framework. Palette never runs your code.
It never calls a model and never holds your model keys. Your pals do the thinking.
One server and a Postgres database are the whole install, on a machine you control.
Designed for the phone in your pocket and the Mac on your desk, with Android and Windows too.
Palette is in private testing. Get early access and be among the first builders to bring their pals in.
Get early access