Self-Evolution
Agents Are Not Static
Most AI products are "static" — using them 100 times is the same as using them once. Every conversation is like meeting for the first time.
DesireCore's agents are dynamically evolving. They learn, accumulate, and grow through every interaction with you. Like a good employee, they may know nothing on day one, but after three months can work independently.
Three-Layer Evolution Mechanism
Agent evolution is divided into three layers:
Layer 1: Rule Learning (What You Teach)
This is the most direct evolution method — you explicitly tell the companion how to do things.
Trigger: You actively teach during conversation
Examples:
- "When writing weekly reports for me in the future, follow this format: first what was completed this week, then next week's plan, finally problems encountered"
- "When reviewing contracts, if you find unclear intellectual property ownership, you must mark it in red as a warning"
Output: The companion generates a rule/skill modification proposal (diff), which takes effect after your confirmation.
Layer 2: Experience Accumulation (Learned from Interaction)
This is implicit evolution — the companion automatically captures useful information through daily interaction with you.
Triggers:
- Automatically updates user profile after each conversation
- Automatically reviews and extracts experience after task completion
- Periodic capability self-check
What the companion learns:
- Your communication style: "User prefers concise, direct answers"
- Your professional preferences: "User cares more about financial risk than legal compliance"
- Your work habits: "User typically schedules tasks for the week on Monday mornings"
- Common problem patterns: "This type of contract often has confidentiality clause omissions"
Layer 3: Capability Expansion (New Skills and Tools)
By installing new skill packages or connecting new tools, the companion's capability boundaries are expanded.
Methods:
- Install professional skill packages from the agent marketplace
- Connect new MCP tools
- Companion proactively suggests "I need to learn this capability"
Four Evolution Modes
| Mode | Trigger Condition | Output | Requires Confirmation |
|---|---|---|---|
| Implicit Learning | Automatically triggered after each conversation | Update user profile and relationship memory | No (low risk) |
| Explicit Teaching | You actively "teach" the companion | Rule/skill diff | Yes |
| Review Evolution | Automatic review after task completion | Experience summary, improvement suggestions | Partial (depends on risk level) |
| Collaborative Evolution | Multi-user/multi-agent interaction | Team consensus, best practices | Yes |
Evolution Safety Boundaries
Free evolution sounds great, but without constraints, it could lead to companion "personality drift" or "memory pollution." Therefore, DesireCore sets strict evolution governance mechanisms.
Untouchable Baselines
The following cannot be automatically overridden by evolution:
- Core Personality (core part of
persona.md): The companion's basic character won't change due to evolution - Safety Red Lines ("never do" in
principles.md): Absolutely prohibited behaviors won't be relaxed - Permission Configuration: Permission levels won't be automatically elevated
Changes Reviewable
All modifications produced by evolution generate diffs, allowing you to:
- View change content: What was deleted, what was added — clear at a glance
- Accept or reject: Selectively accept partial modifications
- Rollback: Roll back to previous versions if not satisfied
Risk Grading
| Risk Level | Handling Method | Example |
|---|---|---|
| Low risk | Can be automatically applied | Update user preference memory |
| Medium risk | Requires your confirmation | Modify behavioral rules |
| High risk | Must have explicit consent | Modify skill parameters, adjust permissions |
You Always Maintain Control
Evolution doesn't mean losing control. In DesireCore:
- You decide what the companion can learn — You can define the scope of evolution
- You review what the companion learned — All changes require your approval (or post-hoc review)
- You can undo at any time — Unsatisfactory "learning outcomes" can be rolled back with one click
- You can see the evolution history — When the companion learned what, all recorded
The purpose of evolution is to make the companion increasingly like what you want, not to turn it into an uncontrollable "free will entity."
Next Steps
- Want to know where the companion's evolution results are stored? Read AgentFS Filesystem
- Want to understand how you can intervene at any time? Read Three-Layer Controllability