Mentis8 / os8

An operating system.
For enduring intelligence.

Durable agents that pursue goals, verify their work, remember what they learn and improve locally. Built around your firm. Running on your hardware.

Private enterprise intelligenceExplore ↓

An operating system for agents

Give it a goal.
It stays with the work.

os8 gives durable agents a place to act, remember and improve. The runtime and inference live on your hardware. Your custom harness connects that capability to your firm.

01Act

Work in the connected environment.

02Check

Observe what actually changed.

03Continue

Keep progress. Adapt when blocked.

Goal-directed work

Progress that survives
the next message.

Agents break goals into checkable steps, act in real environments and inspect the result. When a step fails, they repair or revise the approach while preserving what already worked.

THE MEASURE OF PROGRESSWhat changed
in the world.

A plausible reply is not proof of a completed task.

ExperienceUseful memoryThe next piece of work

Continuity

Keep what the
work teaches.

Agents retain useful knowledge and proven procedures across tasks. What they remember carries its source, so knowledge can be inspected rather than disappearing into a conversation history.

Local improvement

Learn between
the assignments.

Background work consolidates memory and evaluates improvements locally. Proposed changes must pass evaluation before adoption, with rollback available. Foreground goals take priority over background learning.

Proposed improvement
Evaluate against evidence
Adopt a proven improvement · retain rollback

Fireboard research / results

Small models.
Serious capability.

In our agent evaluations, the runtime behind os8 outperformed OpenAI and Anthropic systems using small, older-generation open-source models on a 128 GB consumer machine.

The thesis behind os8: the system around the model matters. Persistent goals, checked progress, useful memory and local learning turn model capability into sustained work.

TEST HARDWARE128GB

Consumer machine.
Small open-source models.
Local execution.

Comparison scope

This comparison concerns our simulated-company agent evaluations, including Claude Opus 5.5. Full benchmark methodology, model configurations and scores are not published here. Results are specific to the tested tasks and configuration; they do not establish superiority across all tasks or guarantee performance for an enterprise installation.

Sovereign by construction

The intelligence lives here.
We have no access to it.

Mentis8 has no remote backend and no access to customer data. Agent memory, local inference and learning stay on customer-owned infrastructure.

The runtime does not require a cloud service. Slack and email connect through your chosen providers; those communication channels have their own network and data paths.

Meet your agents in Slack and email ↗
Go deeperRead the details & qualification notes

An operating system behind the experience.

People interact with durable agents through Slack and email. os8 and the custom harness provide the operating foundation on dedicated enterprise infrastructure.

Built around your organization.

The runtime grounds work in the requesting user, the connected environment and applicable constraints. Agents act incrementally, check results and retain useful experience. Your pilot establishes the enterprise connections, operating scope and acceptance criteria.

Continue exploring

Applications

Applications

Mentis8 / Private briefing

Begin with
your enterprise.

A briefing starts with your workloads, operating requirements and deployment environment.

Pilot-ready. Contact alex@fireboard.ai to discuss your workload and site requirements.