composio: connect your own composio account to use gmail, slack, github and 1,000+ other apps in chat
typefully: write, schedule and publish posts and threads for x, linkedin, threads, bluesky and mastodon
Changelog6 days ago
bring your own api key 🔑
use your own openai, anthropic or google api key for chat on any paid plan: go to settings → api keys, paste your key, test it, save
chats and project chats on that provider's models run on your key, so the model itself costs no zemith credits; your provider bills you directly (cloud runs still use credits)
only tool use is charged in credits, like web search, image generation, deep research or running code
your key unlocks every model from that provider, including ones above your plan, like claude opus
if your key stops working, the chat stops with a clear message. it never quietly switches to your credits. turn a key off or remove it any time to go back to credits
your key is encrypted with secure kms
it is only ever sent straight to openai, anthropic or google, never to anyone else
remove it and it is deleted right away; deleting your account deletes it too
added gemini 4 hydrogen
added gemini 4 hydrogen to the supported model list
added DeepSeek-V4.1-Flash-1017
added DeepSeek-V4.1-Flash-1017 to the supported model list
cloud improvements ☁️
cloud is faster and more reliable on long tasks
Changelog
added gemini 4 carbon
added gemini 4 carbon to the supported model list
Changelog
the new creative studio 🎨
one composer for everything: image, edit, video, remix, avatar and audio, with the price shown next to generate before you spend anything
remix any video: recast new people into a clip, make a photo copy a clip's moves, swap an item, or restyle the whole clip
recreate these: a wall of remixes and trends you can open on the same clip in one tap
studios for real work: director (script to finished film), ad maker, product studio, virtual try-on, ai influencer, design, characters and brand kit
video tools: captions, dubbing into another language, voice change, lipsync, face swap, relight, character swap and batch edit up to 20 photos
music and sound effects in the new audio mode
start from a template: ugc selfies, headshots, product shots and ads come first, plus styles, scenes and camera moves for video
keep editing a result in words ("now make it night"), with every version kept, and share any creation with a public link
prompt enhance, a community feed to recreate what others made, and new models including minimax h3 and flux 3 max
a wider, faster studio page with bigger previews and fewer taps
Changelog
offline model for chat and live mode on mobile 📱
added support for on-device offline models on mobile for both chat and live mode
chat with text and photos or talk in realtime live mode completely offline without an internet connection
on-device models run locally with zero network latency, use no credits, and keep your offline chats stored privately on your phone
added flux 3 to image generation and image editing, with up to 10 reference images and resolutions from 768px to 4k
flux 3 can search the web before generating for more accurate results (on by default, turn it off in advanced settings)
added flux 3 to video generation: text to video, image to video with a first and last frame, storyboards from 3 to 10 images, and continuing an existing video
flux 3 videos are 5 to 20 seconds at 720p or 1080p, with sound included
new draft mode for flux 3 video: a fast, cheaper 720p preview to try a prompt before the full render
Changelog
added minimax m3.1 flash and mistral large 4
added minimax m3.1 flash to the supported model list
added mistral large 4 to the supported model list
Changelog
fork from a summary to save credits 💸
"fork into new chat" now asks how the new chat should start: copy everything, or start from a summary
start from a summary carries over a short summary instead of the full history, so every new message in the new chat uses far fewer credits. great for long chats or when you want to switch models
works from "fork from here" on a reply too: the summary covers the chat up to that reply
the new chat shows where it came from, and you can open the summary to see what was carried over
a progress dialog now shows while the new chat is being prepared
Changelog
added gemini 4 argon and fixed forking project chats
added gemini 4 argon to the supported model list
fixed forking a chat inside a project: the fork now opens in the same project, and older forks show up and open correctly in the project's chat list
Changelog
skills in projects, video to prompt and prompt generator 🎬
skills and plugins now work in project chats, just like in normal chat
new video to prompt tool: upload a video or paste a youtube link and get a prompt you can reuse
new prompt generator tool: describe what you want and get a detailed, ready-to-use prompt
Changelog
added haiku 5.5
added haiku 5.5 to the supported model list
Changelog
move chats and remove default model
added "move to…" in the chat menu to move chats between projects
you can now remove your default model
Changelog
workflow switch node and multiple reference images 🎛️
image editor steps in workflows now use every reference image you connect, up to what the model supports
new switch step: connect all your reference images, name them, and turn each one on or off. no more unplugging and replugging nodes
auto mode lets the ai pick which references your prompt needs, and picks again when the prompt or images change
Changelog
added gpt sol 6.1
added gpt sol 6.1 to the supported model list
Changelog
added sonnet 5.5 and easier reading while replies stream
added sonnet 5.5 to the supported model list
when you send a message, it moves to the top and the reply fills in below, so you can start reading right away instead of chasing the text as it scrolls
tap the ↓ button to jump to the end of a reply
Changelog
v1.2.2 cached input credits 💸
introducing cached input credits. when a model reuses context it has already seen, like earlier messages, files and tool steps in a long chat, those input tokens now bill at a much lower cached rate
save up to 10x on input credits during long chats and multi-step agent runs