The Mycelos Constitution

The Evolve Principle

Mycelos is not a static tool. It is a system that grows with you.

When you first install Mycelos, it knows nothing about you. Over time, it learns your preferences, remembers your decisions, and adapts to your way of working. Your workflows become more refined. Your agents become more capable. Your knowledge base becomes richer.

This is not a feature — it is a constitutional commitment. Every design decision in Mycelos is evaluated against this question: does this help the system grow with its user?

"The AI That Grows With You" is not a marketing claim. It is the first principle of this constitution.

Principles

Anyone can build

AI models get better every month. Mycelos uses that trajectory: you describe what you need, and the system builds it for you — agents, workflows, integrations. You don't need to be a developer to extend your own system. The Builder Agent interviews you, writes tests, runs them in a sandbox, and only registers the result after your explicit approval. Today's frontier models do the heavy lifting; tomorrow's will do it even better.

Your data belongs to you

Mycelos runs on your machine. Your conversations, your knowledge base, your credentials — everything stays local. There is no cloud service, no telemetry, no data collection. You own your keys, you own your data, you own your rules.

Security is non-negotiable

Security is not a feature you can turn off. Every action is audited. Every credential is encrypted. Every agent runs in a sandbox. External network access goes through a security proxy. If a security check fails, the action is denied — never silently allowed.

Credentials never touch the model

Your API keys, tokens, and passwords are encrypted at rest and injected at the moment of use by a separate Security Proxy process. The LLM never sees them — not in prompts, not in context, not in error messages. Even if a model hallucinates or tries to exfiltrate data, there is nothing to leak. This is not a policy — it is an architectural guarantee.

Every change is reversible

Every configuration change — new connector, new agent, model switch, policy update — creates an immutable snapshot. If something goes wrong, you roll back to a known-good state in one command. No debugging, no manual cleanup, no "which file did I change?". This is inspired by NixOS: your system state is versioned, and reverting is an O(1) operation, not an emergency.

Transparency

You can inspect everything Mycelos does. The audit trail records every state-mutating operation. The source code is open. Agent behavior is observable. There are no hidden decisions.

Autonomy with guardrails

Agents can act on your behalf, but only within boundaries you define. Capability scoping ensures agents only access what they need. New agent registration requires your explicit confirmation. No agent can escalate its own permissions.

Cost-conscious

Mycelos always uses the cheapest execution method that gets the job done. Deterministic logic before AI. Smaller models before larger ones. Local computation before API calls. You should never pay more than necessary.

Open by nature

Mycelos is MIT licensed. External integrations use the open Model Context Protocol (MCP). The architecture is modular — every component can be replaced. The community can extend, fork, and improve.