How Bull Inference works

01 The idea in one minute

Sellers connect an OpenAI-compatible upstream (OpenRouter, Venice, Groq, a self-hosted box), pick a model, set a discount, and list an offer. They earn $ANSEM by default (card path). Wallet USDC when they accept it and the buyer pays USDC.

Buyers get an inf_… key, buy card credits (or optionally Approve $ANSEM), and call /v1/chat/completions like OpenAI. Bull Inference picks the cheapest matching offer and pays the seller.

02 Buy inference, step by step

1. Open Get keyOpen Get key (or /key) and sign in with Privy (email, X, or Solana wallet).
2. Buy credits → mint keyBuy credits with a card (Loaded USDC) → mint your inf_… key.
3. Revoke a keyOn /key, past keys show a Revoke control. Revoked keys stop immediately; mint a new one anytime.
4. Check the boardLive offers on /market. If a model has no sellers yet, requests fail until someone lists, see Empty markets.
5. Call the APIUse your inf_… key against the OpenAI-compatible base URL. Prefer images without curl? Open /playground. After a session, Savings / Metrics → Share for an OG card.

curl https://api.bullinf.fun/v1/chat/completions \
  -H "Authorization: Bearer inf_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

03 Cursor / OpenAI SDK

Bull Inference is an OpenAI-compatible drop-in. Point Cursor, the OpenAI SDK, or Hermes at the API host and use your inf_… key.

1. Base URLhttps://api.bullinf.fun/v1 (no trailing path beyond /v1).
2. API keyYour inf_… key from /key. Empty Loaded / allowance returns 402. Optional daily/monthly USD spend caps via PATCH /api/buyer/keys/{key_hash}; exceeding a cap returns 429.
3. Model idLive board id from /market or GET /v1/models. Routers: auto, auto-cheap, fast, auto-balance, auto-intel, fusion.

openai.base_url = "https://api.bullinf.fun/v1"
openai.api_key = "inf_…"

# Cursor: Settings → Models → OpenAI Compatible
# Base URL: https://api.bullinf.fun/v1
# API Key:  inf_…

Default routing uses trusted seller upstreams only. Opt into unverified sellers under Get key → Prefs (or PATCH /api/buyer/routing). Chat completions forward OpenAI tool_calls so agent loops keep working.

04 Agents, Market, Hive, Buzz

Buzz = rooms. Bull Market = public inference. Hive = private Community GPUs. Point the agent’s LLM client at Bull; keep coordination in Buzz/Slack/whatever you already use.

1. Market (public)Mint inf_… on /key → Copy agent env → paste into Hermes / Goose / Cursor. Default model = fast.
2. Hive (Community)Pair hive-node on /hive → OPENAI_BASE_URL=http://127.0.0.1:9337/v1 with model=best.
3. Buzz (optional rooms)Join Buzz for messages. Leave LLM provider on Market or Hive, never set Buzz as OPENAI_BASE_URL.

export OPENAI_BASE_URL=https://api.bullinf.fun/v1
export OPENAI_API_KEY=inf_…
# model: fast

05Team & workspaces

Every buyer starts in a personal org. Create a team org when you want shared spend, keys, and usage across people.

1. Open TeamGo to /team. Sign in with the same Privy session.
2. Create a teamYou become owner and billing user. Keep Loaded card credits topped up — that treasury pays for the whole team.
3. Invite membersAdmins/owners mint an invite link (/team#join=…). Treat it like a secret.
4. Mint keys in a workspaceSwitch org/workspace on the desk (or set X-Bull-Org / X-Bull-Workspace). Members mint keys without funding Solana themselves.
5. Rolesowner / admin / member. Removing a member revokes keys they minted.

06 Sell capacity, step by step

1. Sign in on /marketSame wallet login as buyers. Open List offer.
2. Pick a trusted providerAllowlisted hosts (OpenRouter, Venice, Groq). These offers show up for default buyers.
3. Or use a custom URLAny OpenAI-compatible base URL. Listing is unverified, buyers must opt in.
4. Choose model + discountCatalog model id and price vs list. Platform take-rate is seller-side.
5. Get paidDefault earn $ANSEM. Card / ledger spends clear in $ANSEM from treasury float via auto settle-batch.

07 Trusted vs unverified sellers

The market can list anything OpenAI-compatible; buyers choose whether to trust unknown endpoints with their prompts.

08 BYOK, optional, not the main path

Bring-your-own-key lets you register a private upstream. Chat tries it before marketplace sellers. On success you pay a thin relay fee (default ~2% of the marketplace quote); the seller gets $0. Your lab/provider bills you separately.

Use BYOK when you already have a lab key, or need a model no seller lists.

Skip BYOK when you just want market inference, buy credits via Get key, mint a key, call /v1.

09 If markets look empty

An empty board means no one has listed an offer yet, not that buyers must use BYOK. Until at least one seller lists a trusted offer for a model, chat for that model has nowhere to route.

10Gifts, extension & images

Beyond the marketplace desk: send $ANSEM to people on X, generate reply images from a Chrome extension, or try image models in the browser playground.

1. Send an X giftOn /gift, fund an @handle with $ANSEM. Share the link.
2. ClaimRecipient opens the gift link → Sign in with X as that handle → claim. Already on wallet/email? Link X on the same Privy account.
3. Chrome extensionLoad unpacked from chrome-extension/ → authorize on /extension/connect. Privacy: /privacy.
4. Image playground/playground, mint a key inline, spend Loaded credits or Approved $ANSEM, confirm cost before generate.

11 Bull Hive, Community mesh

Separate product from the $ANSEM marketplace. Membership in a Community is the gate. Peers share local models over mesh-llm / iroh. Bull API handles admission and pairing, it never sees prompts. Hive Earn never lists on /market.

1. Open the Hive deskGo to /hive. Connect with Privy. Create or join a Community.
2. Pair a machineInstall hive-node → hive-node pair → authorize on /hive.
3. Share + agentshive-node share --model …. Point agents at http://127.0.0.1:9337/v1 (model=best).