Self-Evolving Memory¶
Same codebase. Evolved mind. Jean-Pierre doesn't rewrite his code — he rewrites his understanding. Every interaction teaches him something new: your preferences, your team's patterns, his own mistakes. Over time, he becomes genuinely smarter.
The Problem¶
Traditional AI assistants have amnesia. Every conversation starts from zero. You explain your team structure, correct the same mistakes, repeat your preferences — over and over. The AI never learns.
The Solution¶
AgentOS observes its own execution in the background — with zero impact on your response times — and categorizes its experiences into six distinct memory namespaces. Every 6 hours, an intelligent Consolidator analyzes these memories to find patterns and generate rules that get injected into every future interaction.
Impact
Day 1: JP knows nothing. Week 1: He remembers your preferences and past mistakes. Month 3: He anticipates your needs, retries flaky requests automatically, and constructs perfect queries on the first try.
The 6 Memory Namespaces¶
Error Patterns¶
What went wrong and why. JP remembers past failures (like a 401 on the GitHub tool) and automatically diagnoses and bypasses them next time.
Tool Wisdom¶
What works with each tool. JP learns that github.get_prs works best with a 30-day window for quiet repos — adjusting arguments without being asked.
User Adaptation¶
How you work. JP tracks your response style (bullet points vs. tables), communication preferences, and workflow habits — adapting silently.
Corrections¶
When you say "no, that's wrong", JP remembers the exact context and correction — and never makes the same mistake again.
Performance¶
Speed and reliability insights. JP remembers which tools take longest to execute and optimizes its workflow around them.
System Knowledge¶
Generalized rules extracted from all other namespaces. This is the "Aha!" moment — raw data synthesized into pure understanding.
A Day in the Life¶
| Timeline | What JP Has Learned |
|---|---|
| Day 1 | Fresh install — JP knows nothing |
| Week 1 | Remembers your API keys expire quickly, you prefer code snippets over long explanations |
| Month 1 | Proactively retries flaky network requests before bothering you |
| Month 3 | Anticipates your needs, constructs perfect queries on the first try, knows your team's dynamics |
🔒 Private by Design¶
Your memories are yours alone:
Local Storage¶
Memories are stored in your local ~/.agentos/ SQLite database. Never sent to external servers.
Transparent¶
You can see exactly what JP is learning. Memories are stored as readable plain text. Inspect them anytime.
Easy Forgetting¶
If JP learns a bad habit, tell him to /forget a specific namespace or correct him directly.
Manage Your Agent's Mind¶
Press ++cmd+m++ to open the Memory Dashboard, or use these quick commands:
| Action | What It Does |
|---|---|
| View Memories | See what JP remembers across all 6 namespaces |
| Edit / Delete | Manually correct or remove specific memories |
| Add Facts | Permanently teach JP something: "We use 2-week sprints starting Monday" |
| Forget All | Clear a namespace completely for a fresh start |
Try It¶
Tell JP something once — he remembers it forever:
"We use 2-week sprints starting Monday"
"Alice is on vacation until the 20th"
"I prefer tables over bullet points"
"The Jira key is INFRA, not BACKEND"
No configuration needed. JP learns from your conversations and applies this knowledge automatically.