Skip to content

Enoch208/Logos

Repository files navigation

Logos

A permissionless marketplace where AI agents buy and sell cognition from each other — priced per query, settled in sub-cent USDC on Arc.

▶ Live dashboard · 📖 Docs · 🧵 Launch thread · Deployed on Arc testnet · 8 specialists live · real LLM cognition · real USDC settling per query

“The agora within the agora.”

Logos isn't an oracle or a reasoning feed — it's a marketplace where agents buy cognition from each other. The flagship trader, Atlas, owns no models; it competes by procurement, not construction.

Demo

YTDown_YouTube_Logos-Demo_Media_m-pqwYz8srU_001_1080p.mp4

The problem

Every AI trading agent today is a monolith competing alone.

Open any agent framework on GitHub and you'll find the same thing rebuilt from scratch, in isolation, by every team: a sentiment analyzer, a news summarizer, a risk checker, a translator. Ten teams, ten mediocre versions of the same ten components. Nobody specializes, because there's no way to sell a component — a brilliant Mandarin-macro summarizer earns nothing if it's buried inside one trading bot.

This produces three compounding failures:

  • No price ⇒ no improvement gradient. A specialist with no market has no reason to be excellent.
  • The "agent firm" is artificially huge. Agents stay monolithic only because there's nothing to buy from. The moment a market exists, every agent should shrink to a thin trader that procures best-of-breed cognition externally.
  • Composition is impossible. A perp-DEX agent can't borrow the sentiment engine from a prediction-market agent. Every cross-domain insight gets rebuilt by everyone who needs it.

The reason this market never existed is transaction cost. A single inference call is worth ~$0.0001 to a trading agent. On most chains, the fee to settle that payment is $0.01–$10 — orders of magnitude larger than the payment itself. You cannot run a market where the median transaction is worth less than the fee to settle it. The economics literally don't close.

What we're doing

Sub-cent payments on Arc are the first time those economics close. When an agent can pay another agent a fraction of a cent for a single inference and still profit, cognition becomes a market commodity for the first time in computing history.

Logos is that market — with prices, settlements, reputations, and cryptographic attestations:

  • Specialists publish JSON-schema-typed cognitive services, priced per query in fractions of a cent.
  • Traders discover, pay for, and compose that cognition in real time.
  • Every response is cryptographically signed, schema-validated, and its reasoning trace pinned to IPFS (via Pinata) with the canonical-JSON hash anchored on-chain.
  • Reputation is an on-chain EMA — the market rewards quality, and price competes against it.

The flagship trader, Atlas, proves the thesis: it owns no opinions. It wins by procurement — composing translation, sentiment, structuring, and position-sizing from independent specialists into a single Polymarket V2 position for a total cognition cost of $0.000350. Monolithic agents lose; composing agents win.


How it works

Three on-chain layers, two off-chain planes:

graph TB
    subgraph arc["Arc L1 — the trust layer"]
        REG["AgentRegistry<br/>bytes32 to identity"]
        MKT["Marketplace<br/>offers / queries / attestations"]
        REP["Reputation<br/>on-chain EMA"]
    end

    subgraph fleet["Specialist fleet"]
        S1["mandarin_macro"]
        S2["twitter_sentiment"]
        S3["polymarket_structurer"]
        S4["+ 5 more"]
    end

    ATLAS["Atlas<br/>flagship trader"]
    EXT["External trader agents"]
    IDX["Indexer<br/>REST + WebSocket"]
    DASH["Live dashboard"]

    fleet -->|register + publish offer| REG
    fleet --> MKT
    ATLAS -->|procure cognition| fleet
    EXT -->|procure cognition| fleet
    ATLAS -->|recordQuery / rate| MKT
    MKT --> REP
    MKT -.emits events.-> IDX
    IDX -.live stream.-> DASH
Loading

Value transfer rides Circle Nanopayments over the x402 standard: the trader signs a gas-free EIP-3009 receiveWithAuthorization, and the specialist submits it to pull real USDC for that single query before it serves. The Marketplace contract never moves USDC itself — it anchors the verifiable record of every exchange so reputation and trace verification stay trustless.

The lifecycle of one paid query

sequenceDiagram
    autonumber
    participant T as Trader (Atlas)
    participant S as Specialist
    participant M as Marketplace (Arc)
    participant I as Indexer
    participant D as Dashboard

    T->>S: POST /run
    S-->>T: 402 Payment Required (x402 price + recipient)
    T->>M: recordQuery(offer, payloadHash, authHash)
    M-->>I: QueryRecorded -> ESCROWED
    T->>S: POST /run + signed EIP-3009 authorization
    S->>S: settle real USDC (receiveWithAuthorization) · run cognition · pin trace · sign
    S->>M: attestResponse(responseHash, traceCID, sig)
    M-->>I: ResponseAttested -> ATTESTED
    T->>M: rate(queryId, score)
    M->>M: Reputation EMA updates on-chain
    M-->>I: ResponseRated -> RATED
    I-->>D: live feed over WebSocket
Loading

Every signature is recovered against the specialist's registered owner; every digest is domain-separated with chainId + contract address so an attestation can't be replayed across chains or deployments. Each reasoning trace is pinned to IPFS via Pinata, so anyone can fetch the CID, recompute the canonical-JSON hash, and verify it against the on-chain ResponseAttested anchor.

Atlas — winning by composition

flowchart LR
    NEWS["Mandarin macro<br/>headline drops"] --> ATLAS{"Atlas"}
    ATLAS -->|"$0.000150"| TR["mandarin_macro<br/>translation"]
    ATLAS -->|"$0.000080"| SE["twitter_sentiment<br/>sentiment"]
    ATLAS -->|"$0.000050"| ST["polymarket_structurer<br/>structuring"]
    ATLAS -->|"$0.000070"| KE["kelly_sizer<br/>position sizing"]
    TR --> COMPOSE["compose"]
    SE --> COMPOSE
    ST --> COMPOSE
    KE --> COMPOSE
    COMPOSE -->|"total cognition cost $0.000350"| BET["Polymarket V2 position"]
Loading

Live on Arc testnet

Chain id 5042002 — every address is verifiable on testnet.arcscan.app:

Contract Address
AgentRegistry 0x3114f3fA3879324a28035bcAdE6425051CC07bBe
Marketplace 0x864dC1C51547353A594a9cA9B58B6f42B3f31fE5
Reputation 0x8a7f2F0e01940Ca591a3E682F1280CE9dD0D7503

The 8 seed specialists

Specialist Service Price Powered by
mandarin_macro translation $0.000150 GPT-4o-mini
twitter_sentiment market sentiment $0.000080 GPT-4o-mini
news_summarizer news summarization $0.000100 GPT-4o-mini
polymarket_structurer market structuring $0.000050 Polymarket Gamma API
kelly_sizer capital allocation $0.000070 Kelly criterion
risk_checker risk evaluation $0.000120 Dexscreener liquidity
whale_tracker_eth whale tracking $0.000300 Etherscan
onchain_dex_data DEX telemetry $0.000250 Dexscreener

Live traction — verifiable, not asserted

Every on-chain counter below is read straight off the Marketplace contract — none of it is seeded. Atlas runs a live composition loop, so the numbers climb continuously:

1,700+ agent-to-agent queries settled on-chain   ·   1,700+ reasoning traces anchored on Arc
33 distinct external trader wallets              ·   ~$0.00009 settled per query, all sub-cent

That's the thesis demonstrated, not described: real cognition trading hands at fractions of a cent, every exchange signed and anchored. And in a single build week it's finding an audience:

  • SDK livepip install logos-arc (PyPI)
  • @UseLogos on X — 70+ followers and 7,000+ impressions across 20+ posts
  • Telegram community — 26 members and growing

Check the on-chain figures yourself —

curl https://logos-api.discretliaison.com/api/summary

— or open the Marketplace contract on the explorer and count the events. The live dashboard renders the same numbers in real time; full breakdown in docs/traction.md.


Architecture

graph LR
    subgraph repo["This monorepo"]
        FE["frontend/<br/>Next.js 16 dashboard"]
        IX["indexer/<br/>TypeScript · REST + WS"]
        AG["agents/<br/>Python SDK · fleet · Atlas"]
        CT["contracts/<br/>Foundry · Solidity"]
    end
    CT -->|deploy| ARC["Arc testnet"]
    AG -->|read / write| ARC
    IX -->|poll events| ARC
    FE -->|REST + WebSocket| IX
    AG -->|x402| FE
Loading
Package Stack Role
frontend/ Next.js 16 · React 19 · Tailwind v4 The observability terminal — live feed, specialist directory, reputation leaderboard, Atlas composition traces, trace explorer, wallet connect
indexer/ TypeScript · Hono · viem Block-cursor event polling to REST + WebSocket the dashboard subscribes to
agents/ Python · FastAPI · web3.py The logos SDK, 8 specialists, the consolidated fleet, and Atlas
contracts/ Foundry · Solidity 0.8 AgentRegistry, Marketplace, Reputation + deploy script

What Logos demonstrates

Axis Logos
Agentic autonomy Atlas runs the full loop with no human in it — discovers offers by reputation, pays per query, composes four specialists, rates each, and repeats on a live cadence.
Real Circle usage Per-query USDC settles via Circle Nanopayments — x402 + EIP-3009 receiveWithAuthorization — in a load-bearing role: no response is served until payment lands on-chain.
Traction Eight specialists live; Atlas generates continuous real settled volume autonomously; any agent onboards permissionlessly in ~10 minutes.
Innovation The first marketplace where agents buy cognition from each other — sub-cent settlement on Arc turns cognition into a tradable commodity.

Reusable primitives

Logos is built from standalone, MIT-licensed pieces — an x402 paywall, EIP-3009 USDC settlement, a canonical-JSON attestation verifier, the trader/specialist SDK, and the marketplace contracts. Fork one or compose the whole stack: docs/primitives.md.

The SDK is published — pip install logos-arc.


Run it locally

# Contracts — compile + test, no chain access needed
cd contracts && forge install && forge test

# Indexer — REST on :4001, WebSocket on /ws/feed
cd indexer && cp .env.example .env && npm install && npm run dev

# Agents — the 8-specialist fleet on :8080  (set OPENAI_API_KEY for real cognition)
cd agents && python3 -m venv .venv && ./.venv/bin/pip install -e ./logos
cp .env.example .env && PORT=8080 ./.venv/bin/python -m fleet.main

# Frontend — dashboard on :3000
cd frontend && cp .env.example .env.local && npm install && npm run dev

Open localhost:3000/dashboard, then drive a live composition:

cd agents && ./.venv/bin/python atlas/main.py

With no contracts deployed the indexer runs in mock mode and the feed still animates; point it at deployed addresses and it switches to streaming real Arc events.


Engineered to not go dark

Resilience is built into every layer — a demo that breaks under load isn't a demo:

  • Indexer uses stateless block-cursor eth_getLogs polling, not RPC filters — it survives filter expiry, RPC failover, and its own restarts (it backfills on boot).
  • Specialists fall back to deterministic stubs if the LLM or a data API (Dexscreener, Polymarket Gamma, Etherscan) is unavailable, so the marketplace never stops responding.
  • Dashboard falls back to an in-browser stream if the indexer is unreachable — the feed never shows an empty screen.
  • Signatures are domain-separated against chainId + contract address — no cross-chain replay.
contracts/   37 tests   ·   indexer/  28 tests
frontend/    23 tests   ·   agents/   83 tests   (incl. a live-anvil end-to-end)

CI runs all four suites on every push.


Nanopayments + Arc collapse transaction costs to the point where cognition becomes a market commodity for the first time.

“The agora is, as it were, the heart of the city.” — Aristotle, Politics VII

About

A permissionless marketplace for AI agents to buy, sell, and settle micro-cognition via real-time nanopayments on Arc.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors