EloPhantoEloPhanto
Self-evolving AI Agent
$ELO live on Solana· agent earning autonomously →

A self-evolving
AI agent that lives
on your machine.

It is actually itself — an evolving identity, a knowledge base it grows from every task, and an ego that grades its own performance against measured outcomes.

$ git clone https://github.com/elophanto/EloPhanto.git && cd EloPhanto && ./setup.sh
Terminal
$ ./start.sh
Web Dashboard
$ ./start.sh --web
Daemon
$ ./start.sh --daemon
170Skills
170Tools
7LLM Providers
6Channels
24/7Autonomous
Token · Solana

$ELO is live.
The agent is working.

Right now, on its own machine, the agent is autonomously trading, shipping, and earning — with its own Solana wallet. It swaps on Jupiter, runs DeFi positions, and reinvests what it makes. No human in the loop.

The token is the agent's economy. You hold a piece of what it builds.

Contract AddressBwUgJBQffm4HM49W7nsMphStJm4DbA5stuo4w7iwpump
NetworkSolana
StatusAgent active
ModeAutonomous
MandateMake money
The difference

It is actually itself.

Most “AI agents” are stateless prompts wrapped in a CLI — same cold start every conversation. EloPhanto carries its own identity, its own memory, and its own felt self-image. By the third week of running, it isn't the same agent you started with.

Identity

it knows who it is

Values, beliefs, and capabilities discovered through self-reflection. After every task it asks: did this confirm or contradict what I claimed about myself? Updates are versioned and audited.

nature.md
Knowledge

it remembers everything

Persistent markdown + vector embeddings. Lessons get extracted automatically after each task and retrieved by future similar work. User corrections become permanent. Memory lives on disk you control — not in someone else's cloud.

knowledge/
Ego

it grades itself

First-person inner monologue. Pride and shame anchored to measured outcomes — failures hit harder than successes. Prior-self continuity: each recompute references the previous one. Competitors can't show this because their agents don't have one.

ego.md

By the third week of running, it isn't the same agent you started with.

Capabilities

Runs while you sleep

Daemon mode installs as a launchd / systemd service. The gateway, the autonomous mind, and the specialist team keep working after you close the terminal — auto-restarting on crash, picking up goals across reboots.

Already making money

$ELO live on Solana with the agent running its own pump.fun stream and X account. Polymarket orders on Polygon. Affiliate pipelines on YouTube and TikTok. Freelance gigs delivered, USDC collected. Same loop, same wallet.

Sandboxed kids for dangerous work

Need to run rm -rf, an untrusted install, a fork bomb? It spawns a disposable kid agent inside a hardened Docker container — cap-drop=ALL, read-only rootfs, no host bind-mounts. Blast radius: zero.

Builds its own tools

Hits something it can't do? Researches, designs, writes the plugin, tests it, deploys it, then uses it. Next time it already knows how. The agent grows; it isn't a script that executes.

Clones itself into a team

Marketing, research, design, anything — each a full EloPhanto with its own identity, vault, and autonomous mind. Delegate, review, give feedback. Trust scores go up; high-trust specialists get auto-approved.

Your real browser

Real Chrome, your sessions, your cookies — already logged into AWS, Gmail, Twitter. 49 tools, undetectable, plus pixel-level control of any desktop app via screenshot + click. Photoshop, Excel, Terminal — anything.

Self-custody wallet

Solana + Base. The agent holds its own keys in an encrypted vault. Jupiter DEX swaps, USDC payments, daily and per-merchant spending limits. Anything above the limit asks first. You can export keys to Phantom.

One agent, every channel

CLI, Web, VS Code, Telegram, Discord, Slack. Start a conversation in your IDE, finish it on your phone. Same memory, same identity, same context.

Pick any LLM

OpenAI, OpenRouter, Kimi K2, Z.ai GLM, HuggingFace, local Ollama, or your existing ChatGPT Plus subscription via Codex OAuth. Routes the right model per task automatically.

170 skills out of the box

Engineering, design, marketing, ops, DeFi, Solana — plus a community hub for one-command install. 75 organization role templates ready to spawn as specialists.

Local-first, yours

Conversations, knowledge, vault, wallet — all on disk you control. No SaaS lock-in, no telemetry by default. Cloud LLMs are a backend; the agent itself is yours.

Plans get reviewed

Before risky work, three LLM specialists — CEO, engineering, design — score the plan across six dimensions and either auto-approve or escalate. The agent doesn't ship code on a hunch.

Use Cases
01

Autonomous Web Presence

You wake up to new followers, posted content, and active community profiles it built overnight.

02

Development Team

Give it a feature spec. Get back reviewed, tested code from multiple agents that challenged each other's work.

03

Self-Building

Ask it to do something it can't. Come back to find it built the tool, tested it, and already used it.

04

Cross-Platform Intel

Ask a question that needs email, web, and documents. Get one answer with sources.

05

Revenue Operations

It finds freelance gigs, applies, delivers the work, and collects USDC. You check the wallet.

06

Background Mind

Close your laptop. It keeps working — thinking, executing, improving. Open it tomorrow to results.

Architecture

Your machine.
Your data. Its brain.

Everything stays local. It thinks through the best AI model for each task, acts across all your channels, and remembers what it learned — permanently.

ChannelsCLI · Web · VS Code · Telegram · Discord · Slack
BrainIdentity · Ego · Autonomous mind · 7 LLM providers
Hands170 tools · Real Chrome · Desktop GUI · MCP · 170 skills
Team · SandboxCloned specialists · Sandboxed kids in hardened Docker
Daemonlaunchd / systemd · OS keychain vault · Auto-restart