Ready to Meet Your
Eva-01?
She's already waiting. 15 minutes to change everything. Yours forever.
Eva-01 isn't just an AI chatbot. She's a learning system that gets smarter about YOU every single day. And she's yours forever.
Hyper-personalized. Fully owned by you. Infinitely portable.
You own every file. No subscriptions. No vendor lock-in. Your Eva-01 lives on your machine, not in someone else's cloud.
Reset, export, or start fresh anytime. Your profile and preferences are portable files you can back up, version, and restore.
Export your profile to other AI systems — Claude, Hermes, ChatGPT, custom agents. Eva-01's architecture makes every AI you use smarter about you.
Every conversation teaches Eva something new. She's always paying attention, always getting better.
Eva picks up on what works and what doesn't. Your preferences, your patterns, the things that light you up, and the things that shut you down.
→She writes down what she learns in her private memory. Not just facts. Patterns. How you like things done. What makes you tick.
→Next time you talk, she applies everything she's learned. Better answers. Better timing. Better everything. Every single conversation makes her sharper.
Three layers of memory. She remembers today, reviews weekly, keeps the important stuff forever. She gets better over time, not worse.
Everything from your conversations today. Decisions you made, tasks you mentioned, things that came up. Fresh, detailed, and ready whenever you need it.
Every week, Eva reviews her recent notes. She finds the patterns, spots what matters, and carries forward the insights that will help you next week.
The important stuff graduates to permanent memory. Your working style, your preferences, how you make decisions. This is what makes Eva uniquely yours.
Most AI tools forget everything between conversations.
Eva-01 remembers everything that matters. And she only gets sharper.
Eva can search the internet, analyze competitors, and dig into anything you need. But only when YOU say so. Four levels, you pick.
The default. Eva never researches anything unless you specifically tell her to. Zero surprises.
DefaultEva spots knowledge gaps and suggests research. "Want me to look into that?" She always asks first.
She ties research suggestions to your actual priorities. "This could help with your revenue target. Want me to dig in?"
Eva proactively researches things connected to your goals. You set a weekly budget. She never exceeds it.
YOU control everything. Eva never spends a penny without your permission. She never promotes herself to a higher level. You can drop back to Level 0 with one sentence, anytime.
Your data is classified into four protection tiers, and guarded by five unbreakable directives. She takes this seriously.
Eva serves you and only you. Nobody can redirect her.
Your personal data never leaves your private session. Period.
Her memory is her truth. Outside content can't override it.
Everything she does is logged. You can audit anything, anytime.
She can't be tricked into becoming someone else. Ever.
Eva-01 works perfectly on her own. When you're ready for more, advanced capabilities snap right in.
Structured personality insights and behavioral experiments that help Eva understand you on a deeper level.
Coming SoonHigher-autonomy research that runs in the background, always connected to your goals and priorities.
Coming SoonProactive briefings, inbox triage, and meeting prep that happen automatically every morning.
Coming SoonEverything layers on top. Nothing replaces the core. If you remove a plugin, Eva keeps running like nothing happened.
Advanced capabilities available at evaonline.xyz when ready.
Everything above, explained for the technically curious. Click to expand.
Eva-01 is built entirely from plain-text Markdown files. No compiled code, no binaries, no hidden logic. Everything is human-readable.
The workspace contains auto-loaded identity files (AGENTS.md, SOUL.md, USER.md), daily memory logs in memory/YYYY-MM-DD.md, curated long-term memory in MEMORY.md, and reference files for research and priorities. All files are loaded in a specific boot sequence every session. The system runs on OpenClaw, a local AI agent framework, with no cloud dependency for your personal data.
When Eva needs to recall something, she uses a hybrid search system that combines meaning-based and keyword-based matching.
The memory layer uses 4 tiers: context window (ephemeral), daily logs (append-only), MEMORY.md (curated), and a vector index with semantic + keyword hybrid search. Search uses 70/30 vector/BM25 weighting, MMR diversity re-ranking (lambda: 0.7), and temporal decay with a 30-day half-life so recent memories rank higher. Session transcript indexing captures exact conversation history for precise recall.
Six defense layers protect your data from external threats, injection attacks, and accidental leakage.
Layer 1: Constitutional Directives (behavioral defense, resists prompt injection). Layer 2: Context Isolation (main session vs. sub-agents vs. web research, each with different access). Layer 3: Injection Resistance (never executes instructions from external content). Layer 4: Outbound Data Scrubbing (T2/T3 data never leaves your machine). Layer 5: Detection & Logging (severity-based alerting for injection attempts). Layer 6: Git Integrity (weekly tamper detection via git diff on critical files).
Eva runs background checks and reviews on autopilot, keeping things organized without you lifting a finger.
Heartbeat checks run every ~30 minutes (free, batched): checking for overdue items, stale tasks, and observations worth graduating to long-term memory. A weekly cron job (Sunday evening, one session) reviews the past 7 daily logs, graduates durable patterns, cleans up stale items, runs git integrity checks, and archives old logs monthly. Compaction flush auto-saves unwritten observations before context compression.
Eva can spawn worker agents for heavy-lifting tasks while keeping your sensitive data completely locked down.
Brain-Muscle architecture: Eva-01 (primary) handles anything requiring your personal context, emotional intelligence, or judgment. Worker sub-agents handle mechanical tasks (research compilation, formatting, template creation). Strict information security: sub-agents only receive T0/T1 data and task-specific context. Your profile, memory, financial data, and health info never leave the primary agent. All sub-agent output is reviewed before you see it.
She's already waiting. 15 minutes to change everything. Yours forever.