# smolsi

> A complete inference system, running on hardware you own.

smolsi, short for "smol super intelligence," is built by Happiest Labs (Happiest, Inc.). It brings together a local inference engine, a document index, tools, and background workers. The system supports direct answers, vision, and analysis inside the customer's environment, including air-gapped deployments.

This guide reflects the product website as of November 3, 2026. The site is accepting evaluations through its private beta.

## The inference stack

1. **Your silicon:** Hardware selected and sized around the customer's workflow.
2. **smolsi engine:** An in-house inference engine tuned to its hardware, with certified open-weight models, direct answers, and vision in the same session.
3. **Document index:** Documents, spreadsheets, and images become a searchable local index. Relevant material is selected for the model's context.
4. **Tool surface:** The model specifies the analysis; the host executes calculations and renders tables and charts.
5. **Workers:** Background workers handle indexing, curation, and delegated analysis while the foreground session remains responsive.
6. **smolsi, assembled:** These components form one local system on hardware the customer controls.

See the [inference stack](https://smolsi.com/#stack) on the website.

## Air-gapped agent workflows

An agent workflow can run within the customer's private environment, with local documents, local inference, and host-executed tools. An air-gapped deployment does not depend on a hosted model service for this work.

The workflow is:

1. **Bring local documents:** The local index makes relevant material available to the task. Customer source files remain inside the private deployment.
2. **Delegate work:** The local model directs the analysis and delegates work to tools and background workers.
3. **Compute on the host:** The model specifies the calculation or workbook; the host performs the computation and renders the resulting tables and charts.
4. **Return results for review:** Results come back to the user for review and follow-up in the same local workflow.

For example, a user can ask a question over internal documents, have the model specify an analysis of a local spreadsheet, and review the host-computed table or chart. The model directs the work; the host executes the calculation.

See [agent workflows](https://smolsi.com/#agents) on the website.

## Models and capabilities

Local model families include **Qwen, Llama, DeepSeek, Mistral, Gemma, Kimi, and gpt-oss**. Qwen is certified for the current release; additional families require certification before deployment. Ask about the model and configuration you need during your evaluation.

- **Chat and vision:** Direct answers and vision on the local engine, with the customer's information in context.
- **Document index:** Search documents, spreadsheets, and images locally; relevant material enters the model's context.
- **Workbook and tools:** Model-directed analysis with calculations, tables, and charts executed on the host.

See [capabilities](https://smolsi.com/#platform) and the [FAQ](https://smolsi.com/#faq).

## Deployment and licensing

| Configuration | Availability | Intended use |
| --- | --- | --- |
| Workstation | Available | One analyst using a signed desktop application that can work with the network unplugged. |
| Shared node | Available | Local inference for a team, with concurrent sessions and background workers. |
| Private cluster | Early access | Dedicated capacity for isolated environments, with deployment engineering. |

One team evaluates the workflow, configures the system, and supports the deployment. Hardware is sized around the customer's workload; hardware selection and procurement are available through the deployment team.

Licenses are annual, per seat for workstations or per node for shared inference. Enterprise deployments are quoted. There is no per-token metering. Pricing, hardware, and services are discussed during the evaluation.

See [deployment](https://smolsi.com/#deploy) and [licensing](https://smolsi.com/#pricing).

## Privacy and deployment controls

The model, document index, and tools can operate with the network disabled. In an air-gapped deployment, source documents and analysis stay within the configured customer environment.

Deployment controls are configured around organizational requirements for data residency, access control, audit logging, encryption, and network isolation. Document ingestion includes prompt-injection scanning, with the verdict stored alongside the document. Production model directories carry byte-level manifests and release stamps.

Supporting documentation for security reviews is available under NDA. Ask the engineering team about the controls and evidence relevant to a particular deployment.

See [security and assurance](https://smolsi.com/#trust).

## Compliance and assurance

smolsi's compliance and assurance credentials include **HIPAA, SOC 2, GDPR, ITAR, CMMC, GLBA, and NIS2**.

Supporting documentation and the applicable scope are available during security review. The engineering team can provide supporting material under NDA. See [compliance and assurance](https://smolsi.com/#compliance) or contact [hello@smolsi.com](mailto:hello@smolsi.com?subject=Security%20documentation).

## Integration availability

This website does not publish a public inference API, an MCP server, or an agent-framework compatibility matrix. The local agent workflow above describes the product's own engine, tools, and workers. It does not establish compatibility with a particular external agent framework.

Discuss endpoint, tool, and framework requirements with the engineering team during an evaluation.

## Evaluation and contact

Bring one real workflow and the documents or task you want to evaluate. The engineering team sizes a configuration around that work. The website offers a 30-minute evaluation call and states that an engineer replies within one business day.

- [Request an evaluation](https://smolsi.com/#evaluate).
- Email [hello@smolsi.com](mailto:hello@smolsi.com).
- Company: [Happiest Labs](https://happiestlabs.ai).
- Website: [smolsi.com](https://smolsi.com/).