Active guardrails
Rules, workflow context and Company Brain constraints injected into every prompt. The agent can't ignore them because they travel with each model call: tone, scope, forbidden actions and goals, always present.
Minerva Stands is the auditable runtime for governed operating instructions. It connects Company Brain context to agent work with active guardrails, path-level access control and embedded audit logging. Official Docker images for deployment — coming soon.


A Company Brain gives agents the operating instructions. Stands is how those instructions can run with guardrails, workspace limits, review and audit. We start from nano-bot, the autonomous agent from the University of Hong Kong, and harden it with four Minerva layers so it can operate in real environments without surprises. Everything is open source. Official Docker images for deployment are coming soon.
Rules, workflow context and Company Brain constraints injected into every prompt.
Allow/deny list of editable paths.
Every interaction is normalized and stored in a SQLite database embedded in the agent itself.
Built-in UI to inspect the workspace, chat with the agent, validate intermediate actions, and query the SQLite audit log — all in one place.
Agents should not improvise from raw documents. Stands keeps model actions inside reviewed instructions, allowed paths and an audit trail.
Rules, workflow context and Company Brain constraints injected into every prompt. The agent can't ignore them because they travel with each model call: tone, scope, forbidden actions and goals, always present.
Allow/deny list of editable paths. You define what the agent can read, write or ignore, so Company Brain instructions cannot become broad filesystem access.
Every interaction is normalized and stored in a SQLite database embedded in the agent itself. Full traceability: Company Brain context, prompts, responses, tools used and file changes.
Built-in UI to inspect the workspace, chat with the agent, validate intermediate actions, and query the SQLite audit log — all in one place.
Three production deployments with Minerva customers—not slideware. Each agent owns a workflow and acts from governed context: calendar and inbox, social content, or customer memory.
A digital secretary with access to the calendar and inbox. Prioritizes and categorizes emails based on whether they need action, oversight, are blocking something, or are purely informational.
Once a day it scrapes news relevant to the company, extracts topics with cited sources, and after user approval writes, reviews and schedules content across social networks.
Active memory with a custom protocol and Minerva's internal tooling. Categorizes and keeps every customer interaction, demo and commitment up to date, wired into Google Calendar.
We believe autonomous agents should only run when the knowledge, instructions and actions can be audited. Minerva Stands is open source end-to-end. Official Docker images are coming soon so you can inspect, modify, and choose where it runs.
View on GitHubTell us the process you want to automate. In 30 minutes we'll tell you whether it needs better knowledge, stronger governance, or an auditable Stands workflow.