A practical start
to local AI.
This documentation is a planning skeleton for the LOCAL AI product family. It explains what to consider now; release-specific instructions will arrive with the software.
Choose your starting point
Installation
Installers, platform support and integrity-verification instructions will be published at early access. Do not treat interface previews as downloadable software.
Prepare your environment
- Record your operating system and available memory.
- Check available storage for model files and workspaces.
- Identify the person responsible for updates and recovery.
- For teams, agree network access and account policies first.
We will document supported installation paths and how to verify release files before asking anyone to install a runtime.
Models
A runtime and a model are different parts of the system. The runtime executes inference; the model provides the learned weights. Model formats, runtime support and licences must align.
Review before use
- Read the original licence and any acceptable-use restrictions.
- Confirm the model source and version.
- Evaluate with representative, non-sensitive tasks.
- Review generated output before relying on it.
LOCAL AI Hub is planned to help with discovery. It will not replace your review of the model’s original terms.
Hardware guide
Hardware fit depends on the model, its format, context length and workload. This preview makes no benchmark or performance claims.
| Resource | What to consider |
|---|---|
| Memory | Model weights, working context and other applications share available memory. |
| Compute | Runtime support determines whether your processor or accelerator can be used. |
| Storage | Keep room for model files, updates and your own workspace data. |
| Operations | Plan power, cooling, updates and recovery for sustained team workloads. |
Supported configurations and evaluation guidance will be published before release. Start with a task and the hardware you already have.
API (planned)
LOCAL AI Server is intended to provide self-hosted inference for teams. The API specification, authentication approach and compatibility details have not been published.
access policy
local inference runtime
response for review
This is a conceptual flow, not executable code. There is no API endpoint on this website. Client examples and error-handling guidance will accompany a released specification.
Security
Local inference can reduce external data exposure, but it does not secure a machine by itself. Treat models, tools and generated output as separate trust boundaries.
- Restrict network access and use least-privilege accounts.
- Review model provenance and release integrity.
- Keep sensitive files outside unapproved agent workspaces.
- Document updates, backups and incident response.
Frequently asked questions
Can I download LOCAL AI today?
No. Studio, Server, Agents and Hub are planned products. Early access and supported platforms will be announced when ready.
Is the chat running a local model?
No. It streams scripted responses in your browser. It does not load a model, access files or send your message to a server.
Will offline use need an initial download?
Yes, the planned offline workflow needs a runtime and model files on your hardware first. Downloads and software updates need connectivity.
Do I need LOCAL AI COIN?
No coin is required for this preview. LOCAL AI COIN is planned, not issued, and nothing here is for sale. Product access terms will be published at launch.
Where can I ask a question?
Email hello@localai.best with your use case. Do not include confidential data in an initial email.