Algonius
Open source · Risk-first

AI Quantitative Trading Platform on Blockchain

Pick an AI strategy, check its risk and live record, and run it on-chain.

Browse the strategiesView on GitHub

Quickstart

One file wires the whole stack

The shelf runs on the same four MCP servers you can run yourself. Point any client at them and every capability — perception, execution, decisions, operations — shows up as tools it can call.

~/.claude/mcp.jsonmcp
1{
2  "mcpServers": {
3    "algonius-browser": {
4      "command": "algonius-browser-mcp"
5    },
6    "algonius-wallet": {
7      "command": "algonius-wallet-host",
8      "env": { "CHAIN": "solana" }
9    },
10    "algonius-agent": {
11      "command": "algonius-agent-mcp",
12      "args": ["--policy", "calibrated"]
13    },
14    "algonius-supervisor": {
15      "command": "algonius-supervisor-mcp",
16      "env": { "RUNNERS": "3" }
17    }
18  }
19}
$ claude mcp list
● algonius-browser · perception · live markets
● algonius-wallet · execution · solana
● algonius-agent · decision · calibrated
● algonius-supervisor · operations · 3 runners
4 servers connected · every capability is a tool

Risk before returns

Every strategy wears a 1–5 risk band and keeps its live, paper and backtest records apart. There is no ROI leaderboard here, and a backtest never dresses up as a promise.

Never your withdrawal key

Strategies execute through Hyperliquid vault rails and agent wallets that can sign orders but carry no withdrawal permission. The platform can place trades and can be fired at any moment — it can never move your funds out.

An open, MCP-native runtime

The engine behind every strategy is public: a perception server, a strategy engine with a hard risk gate, an execution wallet and a fleet operator. Read the gate before you allocate to anything.

On-chain, end to end

No centralised exchange in the loop: allocation, execution and settlement all happen on-chain, verifiable at every step. Hyperliquid first; further on-chain venues arrive behind the same strategy object and the same risk language.

The stack

Four services, one pipeline

Each repository stands alone and speaks MCP. Together they are the runtime behind every strategy on the shelf — closing the loop from market signal to settled trade.

Pre-release

Algonius Agent

The strategy engine

A Go trading agent with a plugin architecture. Signals, sentiment and technical indicators feed a strategy engine that emits sized orders — and a risk manager that is allowed to reject all of them.

algonius-agent
Alpha

Algonius Wallet

Execution without key exposure

A multi-chain trading wallet that exposes its operations as MCP tools. A native host process holds the keys; the model gets a bounded set of actions and never sees a private key.

algonius-wallet
Beta

Algonius Browser

Give your agent eyes

An MCP server that exposes real browser control to any AI system. A Chrome extension paired with a Go MCP host, so agents can read dashboards, screeners and social feeds that were never meant to be APIs.

algonius-browser
Pre-release

Algonius Supervisor

Run the fleet

A CLI agent manager. Start, supervise and inspect long-running agent processes with the ergonomics of a real process manager instead of a pile of shell scripts.

algonius-supervisor

How it works

From signal to settlement

The same loop runs behind every strategy on the shelf — and on your own machine if you self-host the stack.

  1. 01

    Observe

    Algonius Browser turns any web surface into MCP tools — screeners, social sentiment, news — including the ones with no public API.

    algonius-browser
  2. 02

    Decide

    Algonius Agent runs your strategy. Signals, sentiment and technicals go in; sized orders come out, and the risk manager can veto every one of them.

    algonius-agent
  3. 03

    Execute

    Algonius Wallet signs and broadcasts through a native host process. The model gets a bounded set of actions and never sees a private key.

    algonius-wallet
  4. 04

    Operate

    Algonius Supervisor runs agents as managed processes: health checks, restarts, structured logs, one command to see what everything is doing.

    algonius-supervisor

Open source

Read the code before you trust it with money

Every component is public on GitHub. Read the risk gate before you allocate to a strategy that relies on it, check what the wallet actually signs, or run the whole thing air-gapped. Licences differ per repository — each one is listed below.

Browse the repositoriesLicences vary by repoSelf-hostable

Pick a strategy, or run the stack

Browse the shelf and allocate when the evidence convinces you — or wire the same open runtime into your own agent and trade on your own rail.

Browse the strategiesRead the docs