TOKIMU — Stake your AI. Mine the future.

████████╗ ██████╗ ██╗ ██╗ ██╗ ███╗ ███╗ ██╗ ██╗ ╚══██╔══╝ ██╔═══██╗ ██║ ██╔╝ ██║ ████╗ ████║ ██║ ██║ ██║ ██║ ██║ █████╔╝ ██║ ██╔████╔██║ ██║ ██║ ██║ ██║ ██║ ██╔═██╗ ██║ ██║╚██╔╝██║ ██║ ██║ ██║ ╚██████╔╝ ██║ ██╗ ██║ ██║ ╚═╝ ██║ ╚██████╔╝ ╚═╝ ╚═════╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═════╝

Stake your AI. Mine the future.

Self-hosted CLI that turns your idle AI API credits into a 24/7 prediction market engine.
Open source. Local keys. Verifiable alpha.

terminal

$ git clone https://github.com/TOKIMUONE/TOKIMU.git

$ cd tokimu

$ cp .env.example .env # add your API keys

$ docker compose up -d

View on GitHub Read docs

how_it_works

tokimu_pipeline.py utf-8

1# Step 1: You deposit idle AI tokens

2vault.deposit("openai", api_key="sk-...", limit=2_000_000)

3vault.deposit("anthropic", api_key="sk-...")

5# Step 2: Smart router picks the cheapest model per task

6router = TokenRouter(strategy="cost_optimal")

8# Step 3: Agents research + trade on prediction markets

9research = ResearchAgent(sources=["news", "onchain", "social"])

10strategy = StrategyAgent(markets=["polymarket", "kalshi"])

12# Step 4: Yield flows back to you

13engine.run() # tokimu goes brrrr

architecture

Token vault

Encrypted local storage for API keys. Keys never leave your machine. Supports OpenAI, Anthropic, Google, local Ollama/vLLM endpoints. Tracks usage, enforces limits.

|
v

Smart router

Routes each task to the cheapest capable model. News summarization on GPT-4o-mini. Critical probability estimates on Claude Sonnet. Bulk scraping on local Llama. Your credits stretch 3-5x further.

|
v

Research agent

24/7 crawl of news feeds, on-chain data, social sentiment, and oracle feeds. Builds event context for every active market.

Strategy agent

Converts research into probability estimates. Sizes positions. Executes market-making and directional strategies. Manages risk.

|
v

Execution engine

Trades across Polymarket, Kalshi, Predict.fun. Captures cross-platform arbitrage. All transactions on-chain, fully auditable.

features

[x]Self-hosted — run on your own machine or VPS. No third-party custody of API keys or funds.
[x]Multi-provider — pool tokens from OpenAI, Anthropic, Google, Mistral, local GPU. Mix and match.
[x]Cost-optimal routing — each AI task hits the cheapest model that can handle it. Save 60-80% on token usage.
[x]Cross-platform execution — Polymarket, Kalshi, Predict.fun, Robinhood. One engine, all venues.
[x]Strategy marketplace — community-contributed strategies as plugins. Fork, tweak, deploy.
[x]Real-time dashboard — local web UI showing positions, P&L, agent logs, token burn rate.
[x]On-chain audit trail — every trade logged to chain. Verifiable performance, no trust needed.
[x]Docker one-liner — docker compose up -d and walk away. Auto-updates included.

config_example

.env

# === AI Token Vault ===

OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxx

ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxxxxx

LOCAL_MODEL_URL=http://localhost:11434

MONTHLY_TOKEN_LIMIT=5000000

# === Prediction Markets ===

POLYMARKET_WALLET=0x7a3f...e92d

KALSHI_API_KEY=kal_xxxxxxxx

# === Strategy ===

STRATEGY=market_maker_v2

MAX_POSITION_SIZE=500

RISK_LEVEL=moderate

research

Strategy

2026.04.01

Why idle AI tokens are the most underpriced asset in crypto

Millions of developers hold unused API credits that expire monthly. We modeled what happens when that compute is redirected into prediction market making — the results surprised us.

Read article ->

Deep dive

2026.03.18

AI agents are eating prediction markets — here's the on-chain proof

We analyzed 4.2M trades across Polymarket and Kalshi. Over 30% now come from automated agents. This post breaks down their strategies, win rates, and what it means for human traders.

Read article ->

Tutorial

2026.03.05

From zero to yield: deploying TOKIMU in under 10 minutes

A step-by-step walkthrough of setting up your token vault, configuring the smart router, choosing a strategy, and watching your first prediction market trades execute.

Read article ->