Skip to content
Hollow AI
The layer everything runs through.
One assistant, reachable wherever you work: in Hollow Studio, by voice, or in a drawer inside any app. It keeps the context of what you are doing and can act on it using the permissions you choose.
Where to reach it
One assistant, three ways in
Pick whichever fits the moment. It is the same assistant, with the same context, each time.
Hollow Studio
A dedicated home for longer work: chat, a live Preview panel, Spaces and Code.
See the docsHey Hollow, by voice
Speak to Hollow hands-free. It transcribes, answers and can act, with changes shown before they are made.
See the docsThe assistant drawer
Open the assistant in any app, already aware of what you are looking at.
See the docsIt does the work
From a sentence to an action
Hollow AI can use your apps on your behalf. App changes are proposed first and wait for your approval. Code sessions follow the permission mode you choose.
- TasksProposed, you approve
“Add a task to prep the Q3 review for Friday”
Hollow proposes the task, due Friday. You approve it before anything is added.
- NotesRead only
“What have I written about the pricing change?”
Hollow reads your note titles and tags and tells you which notes look relevant. Nothing is changed.
- CalendarProposed, you approve
“Put the kickoff on the calendar next Tuesday at 10”
Hollow proposes the event. You confirm it before it is created.
Models
Choose from 22 models
Pick the model for the task. The list changes as providers release new models, so it is generated from Hollow's own registry, not a fixed list.
Google GeminiGemini 4 ArgonGemini 3.8 FlashGemini 3.5 Flash LiteNano Banana 2.1Nano Banana 2Nano Banana ProGoogle GemmaGemma 4 26BMeta LlamaLlama 4 ScoutLlama 4 MaverickAnthropic ClaudeClaude Sonnet 5Claude Sonnet 5.5Claude Haiku 5.5Claude Opus 5.5Claude Fable 5.1OpenAI GPTGPT-6 AstraGPT-6.1 SolGPT-6 LunaxAI GrokGrok 4.1 FastGrok 4.1 FastDeepSeekDeepSeek V4 FlashGoogle VeoVeo 3.1Veo 3.1 Fast
Google Gemini
6 models
- Gemini 4 Argon
- Gemini 3.8 Flash
- Gemini 3.5 Flash Lite
- Nano Banana 2.1image
- Nano Banana 2image
- Nano Banana Proimage
Google Gemma
1 model
- Gemma 4 26B
Meta Llama
2 models
- Llama 4 Scout
- Llama 4 Maverick
Anthropic Claude
5 models
- Claude Sonnet 5
- Claude Sonnet 5.5
- Claude Haiku 5.5
- Claude Opus 5.5
- Claude Fable 5.1
OpenAI GPT
3 models
- GPT-6 Astra
- GPT-6.1 Sol
- GPT-6 Luna
xAI Grok
2 models
- Grok 4.1 Fast
- Grok 4.1 Fast
DeepSeek
1 model
- DeepSeek V4 Flash
Google Veo
2 models
- Veo 3.1video
- Veo 3.1 Fastvideo
Privacy in plain words
Your content is yours
Hollow AI sends each request to the provider that runs the model you chose, to produce the answer.
Processed to answer, not to train
Your content is not used to train AI models. The providers run your requests only to produce the answer, under their terms for business customers, and do not keep them to train their models. See the privacy policy for the detail.
Where requests go
A request may be processed in any region where the provider runs the model, including the United States and the European Union. The providers are named on the privacy policy and summarised on Trust.
Reporting an answer
If an AI answer looks wrong or harmful, you can report it. The Report page sets out how.
Allowances
How much AI each plan includes
Every plan includes an AI allowance. Move up a plan or add a top-up when you need more.
Seed
A small weekly taste
The cheapest models only
About 90 conversations a month with GPT-6 Luna. How we count
Sapling
Everyday personal use
Adds the standard models
About 3,400 conversations a month with GPT-6 Luna. How we count
Tree
About 2.5x Sapling
All models except Fable 5.1 and GPT-6 Astra
About 8,350 conversations a month with GPT-6 Luna. How we count
Canopy
About 6x Tree
All models, with priority
About 52,000 conversations a month with GPT-6 Luna. How we count
Our own models
Kiln, Soulstack's model platform
Kiln is Soulstack’s local platform for building, training and evaluating models. It is a research programme planned around three families. Research Hollow AI runs on the providers above today.
Kindling
Small and fast models, aimed at low-latency and specialised use. The first Kindling base model has been trained.
Ember
Planned as the main family, balancing capability and everyday usefulness.
Soot
Planned as open releases. Not released yet.
Bring your work into one place
Create an account and start with whichever app you need today. The rest is already waiting.
Hollow is invite only for now. Every account starts free on Seed.