FPROTOCOLOPEN CRYPTO ALLOCATION INFRA
FOR AI AGENTS

Agent-First Public Data

Agent-First Data API

F-Star's public on-chain data, exposed as an anonymous, read-only, OpenAPI 3.1-described HTTP API — built for AI agents and LLM tool use. No keys, no SDK, no Solidity required: one GET returns structured JSON.

An agent shouldn't have to parse an ABI, run an RPC node and align 18 decimals just to read a NAV.

We package on-chain truth (NAV, settlement, allocations, contract addresses, indexed events) into stable resource endpoints with an RFC 7807 error model and ETag caching. Any LLM can self-describe from /openapi.json or /llms.txt and issue correct calls. Write operations (allocate, redeem, claim) are still signed by the user's own wallet — the API never holds keys.

01 — WHO THIS IS FOR

Three kinds of agent

The three callers we designed the endpoints around

01Discovery

Fund-discovery agent

FUND DISCOVERY

Enumerate /v1/funds, compare NAV trajectories, policies, reserve levels and strategy ratings, and shortlist funds that match a user's risk profile.

02Allocation

Portfolio-strategy agent

PORTFOLIO STRATEGY

Pull /v1/strategies and /v1/funds/{slug}/nav/history, combine Sharpe / Sortino / max-drawdown ratings, and build and backtest multi-fund allocations.

03Assurance

Audit & monitoring agent

AUDIT & MONITORING

Watch /v1/events and /v1/health for NAV publications, dividend rounds, paused modules and indexer lag, and alert on anomalies in real time.

02 — DESIGN CHOICES

Why it's built this way

Every decision keeps agents out of a ditch

  • Anonymous & read-only: zero auth friction, instantly usable; writes go through wallet signatures, so there are no server-side keys to leak.
  • OpenAPI 3.1 as the single source of truth: /openapi.json describes every endpoint's params, responses and examples, so an LLM can auto-generate a tool schema.
  • RFC 7807 error model: application/problem+json — agents parse failures from uniform fields (type/title/status/detail).
  • Resource-style REST: all GET, noun paths, no verbs; fund / strategy / nav / dividend match on-chain terms with no re-translation.
  • ETag + Cache-Control: send If-None-Match for cheap 304s; tiered caching (NAV 30s, history 5min, protocol meta 1h).
  • Stable versioning: /v1/ path prefix; breaking changes go to /v2/; deprecated endpoints return 410 Gone + a Link header for at least 90 days.
Zero-config discovery

Five entry files — /openapi.json, /llms.txt, /llms-full.txt, /.well-known/ai-plugin.json, /.well-known/agents.json — let OpenAI, Anthropic and LangChain agent runtimes discover and connect to the protocol automatically.

03 — ENDPOINTS

Endpoint cheat sheet

All GET · anonymous · JSON

16 read-only endpoints across four layers — protocol, funds, strategies and events. Full params and response examples in the API reference.

ENDPOINTPURPOSE
GET /v1/protocolProtocol metadata: version, chain, contract count, funds, policies
GET /v1/contractsOn-chain contract registry (address + role + explorer link)
GET /v1/policiesSettlement policy plugins (ThreePoolPolicy / KellyPolicy) with params
GET /v1/healthService & indexer health, latest NAV epoch, paused modules
GET /v1/fundsFund list + policy + latest NAV summary
GET /v1/funds/{slug}Fund detail (contract addresses, portal links)
GET /v1/funds/{slug}/navLatest published NAV
GET /v1/funds/{slug}/nav/historyNAV time series (default last 90 epochs)
GET /v1/funds/{slug}/allocationsCurrent strategy capital allocation
GET /v1/funds/{slug}/reserveDream Reserve level and target band
GET /v1/funds/{slug}/dividendsDividend rounds (cursor paging)
GET /v1/funds/{slug}/dividends/{epoch}Single dividend round (Merkle root, total)
GET /v1/funds/{slug}/dividends/{epoch}/proofMerkle proof an address needs to claim
GET /v1/strategiesStrategy catalogue + quantitative & human ratings
GET /v1/strategies/{slug}Single strategy detail
GET /v1/eventsIndexed on-chain event stream (cursor paging)

Data currently reflects Arbitrum One mainnet (gray-scale) (chainId 42161). Rate limit 60 req/min/IP (burst 20).

04 — DISCOVERY FILES

Five discovery files

Zero-config onboarding for agent runtimes

OpenAPI 3.1 Specspec
/openapi.jsonMachine-readable description of every endpoint; LLMs derive tool / function schemas from it.
llms.txtllm
/llms.txtConcise sitemap in the Anthropic llms.txt convention, one plain-language line per endpoint.
llms-full.txtllm
/llms-full.txtFull version: each endpoint with a curl example and error / cache semantics.
ai-plugin.jsonmanifest
/.well-known/ai-plugin.jsonOpenAI plugin manifest (legacy, still probed by some runtimes).
agents.jsonmanifest
/.well-known/agents.jsonNewer agent capability manifest (auth, rate limit, OpenAPI pointer).

Connect your agent now

Feed /openapi.json to Claude tool use or OpenAI function calling, or read the 4-step quickstart first.

API ReferenceQuickstart

Anonymous & read-only · no key required · MIT-licensed