smista.ai 0.1.0-alpha.1 is here

The first alpha is here

Today, we are releasing smista 0.1.0-alpha.1, the first version of smista.ai that you can install and use.

This release turns the ideas described in the previous development updates into an end-to-end product: one binary containing an interactive CLI and a local router, with deterministic model selection, streamed execution, tool approvals, session storage and support for both remote and local providers.

It is an alpha, and that word matters. Some interfaces will change, there will be rough edges, and this is not yet the version to place at the centre of a production-critical workflow.

It has also barely been tested in the wild, so you should not expect everything to work perfectly yet. Bugs, platform-specific issues and unexpected behaviour are all possible at this stage.

But it is ready for people who want to try the project, understand how policy-driven model routing feels in practice and help us shape what comes next.

What you can do with this release

smista.ai is built around a simple idea: you should dispatch the job instead of choosing a model every time.

With this first alpha, you can:

• describe a task from an interactive terminal or run it directly from the shell;
• classify prompts and route them through deterministic, human-readable policies;
• use remote providers and local Ollama models in the same workflow;
• preview a routing decision before contacting a model;
• control file and shell tools through explicit permissions and approvals;
• keep provider credentials outside configuration files;
• resume workspace-scoped sessions;
• run the CLI and local router from the same smista binary.

The important part is not only that smista selects a model. It can also explain the decision: which intent was detected, which rule matched, which provider and model were selected, what context was included and which permissions are required.

This is the first usable version of the workflow we set out to build.

Install smista

On macOS and Linux, run:

curl -sSLf https://smista.ai/install/install.sh | sh

The script detects your platform, verifies the download checksum and uses Homebrew when it is available.

On Windows, open PowerShell and run:

irm https://smista.ai/install/install.ps1 | iex

The installer downloads the latest release and adds smista to your user PATH.

You can also build and install smista from source with Cargo:

cargo install smista --locked

This method requires the Rust toolchain.

Whichever method you choose, verify the installation with:

smista --version

You should see smista 0.1.0-alpha.1. The commands are also available on the installation page.

Get started in a few minutes

Open the project where you want to use smista and create its configuration:

smista config init

This creates .smista/config.toml, where you define providers, routing, privacy and tool permissions.

For example, a minimal remote-provider configuration looks like this:

[providers.openai]
type = "openai"
 
[routing.default]
model = "openai/gpt-5.5-mini"

Store the provider credential locally, then validate the configuration:

smista credentials set openai YOUR_API_KEY
smista config check project

Start the local router, check that it is ready and log in:

smista start
smista status
smista login

You can now open the interactive CLI:

smista

Or dispatch one task directly:

smista "review the authentication middleware"

If you prefer a local model, smista also supports Ollama. The complete Get Started guide explains how to configure remote providers or Ollama, define routing policies and preview the route before executing a task.

What comes after the first alpha

This release is a starting point, not a finish line.

The next releases will focus on feedback from real workflows, a smoother terminal experience, more complete usage and trace views, stronger plan and diff review flows and the fixes that inevitably follow the first public alpha.

If something feels unclear, fragile or unnecessarily complicated, we want to know. Open an issue in the smista.ai repository or reach for us on Discord, tell us what you tried and what you expected to happen.

For the first time, smista.ai is no longer only a specification, an architecture or a sequence of development updates.

You can install it, configure a policy and dispatch the job.