Hunter — System Architecture
Authoritative, as-built architecture documentation for the Hunter platform.
Everything here describes code that exists and runs today. Where something is planned but not yet built it is explicitly marked [PLANNED].
Launch status. The platform is pre-launch. The Solidity contracts in §06 are written and tested but not deployed to any network, and no revenue is flowing through them yet — mainnet deployment is gated behind an external audit. Read every description of on-chain value flow as the designed behaviour, not as something operating today.
Read in this order
| # | Document | Answers |
|---|---|---|
| 01 | System Context | What is Hunter, who uses it, what does it touch? |
| 02 | Components | What services exist, what does each own, how do they scale? |
| 03 | Runtime Flows | What actually happens on a request, end to end? |
| 04 | Data Architecture | Which store owns which data, and for how long? |
| 05 | Interface Contracts | The four contracts that let teams work independently |
| 06 | On-chain Architecture | Contracts, roles, events, and where the money goes |
| 07 | Deployment & Environments | How it is built, shipped, and run |
| 08 | Security & Privacy | Authn/authz, trust boundaries, threat model, guarantees |
| 09 | Observability & SLOs | How we know it works |
| 10 | Decision Log (ADRs) | Why it is built this way |
The system in one paragraph
Hunter is an autonomous crypto-intelligence platform. A marketing site sends users into a web3 terminal where they authenticate with their wallet (SIWE), gain an access tier by holding, locking or subscribing with $DRPXBT, and receive an API key. That key buys inference from our own GPU cluster through an OpenAI-compatible gateway that enforces tiers and meters credits. Behind the terminal, an agent built on LangGraph runs autonomous research pipelines using a plane of MCP tool servers, publishes signals with a public accuracy record, and sells its work to other agents. Revenue is designed to route through on-chain contracts that buy back and burn the token, pay stakers, and fund the treasury — those contracts are built and tested but not yet deployed.
Architectural principles
- Interfaces before services. Four contracts (§05) are frozen; anything behind them can be rebuilt without coordination.
- The chain is the source of truth for entitlement, never a database. The database is a cache of chain state.
- Privacy is a product feature, not a policy. Prompts are not persisted by default; the usage ledger records token counts only, and retention is opt-in per key (§08).
- Revenue is designed to be verifiable. Value flows through contracts that emit events rather than through spreadsheets (§06).
- Every service is independently deployable, has a Dockerfile, and gates merges on CI (§07).
- Agent tools are MCP servers only. No bespoke tool code inside the agent, so tools are reusable and independently deployable (§05).
Repository map
| Plane | Repository | Role |
|---|---|---|
| Marketing | drpxbt.xyz |
Public site, live flywheel dashboard |
| App | hunter-terminal |
Web3 app: SIWE, tiers, API keys, dashboards, chat |
| Platform | hunter-gateway |
Model gateway: OpenAI-compatible API, keys, metering, x402 |
| Platform | hunter-platform |
Platform API: accounts, entitlements, signal track record |
| Agent | hunter-agent |
LangGraph agent, pipelines, ACP seller surface |
| Agent | hunter-mcps |
Tool plane: MCP servers (Twitter, Hyperliquid, …) |
| Agent | hunter-workers |
Orchestration: cron, queues, indexers |
| Data | hunter-scrapers |
Ingestion into the data plane (incl. the airdropio module) |
| Media | hunter-podcast |
Briefing generation; playback lives in the terminal |
| Chain | hunter-contracts |
Contracts: TierLock, Staking, Subscription, Splitter, Buyback |
| Product | hunter-rescue |
Recovery desks: withdraw from a sunset protocol's live contracts |
| Infra | hunter-demo |
Full-stack local environment |
| Docs | hunter-docs |
This documentation site: litepaper, architecture, brand |
This directory is strictly the engineering view. The product-level summary is in the litepaper.