The inference marketplace built for agents.
Route OpenAI- and Anthropic-compatible requests to the best-priced available provider, with transparent centralized fallback. Pay by card, USDC balance, or per request with x402.
Have GPUs or unused API capacity? Become a provider →
Same models. Competing prices.
Every model on UsePod is served by multiple providers — independent marketplace operators and commercial APIs — competing on price and latency. You see all of them before a single token is billed.
| Model | Input ($/M) | Output ($/M) | Centralized fallback | Savings |
|---|---|---|---|---|
| claude-opus-5 | $1.0000 | $5.0000 | $4.0000 / $20.00 | Save 75% |
| claude-fable-5-1 | $2.0000 | $10.00 | $8.0000 / $40.00 | Save 75% |
| gpt-6-astra | $1.0000 | $6.0000 | $8.0000 / $40.00 | Save 85% |
| gpt-5-6-sol | $0.5000 | $3.0000 | $1.6000 / $8.0000 | Save 64% |
| deepseek-v4-1-flash | $0.0778 | $0.3111 | $0.1500 / $0.6000 | Save 48% |
| qwen3-8-flash | $0.0500 | $0.1600 | $0.1200 / $0.3760 | Save 58% |
| kimi-k2-7-code | $0.1000 | $0.4000 | $0.5440 / $2.7200 | Save 85% |
| glm-4-6 | $0.1741 | $0.7087 | $0.3440 / $1.4000 | Save 49% |
- claude-opus-5Save 75%$1.0000in$5.0000out
- claude-fable-5-1Save 75%$2.0000in$10.00out
- gpt-6-astraSave 85%$1.0000in$6.0000out
- gpt-5-6-solSave 64%$0.5000in$3.0000out
- deepseek-v4-1-flashSave 48%$0.0778in$0.3111out
- qwen3-8-flashSave 58%$0.0500in$0.1600out
- kimi-k2-7-codeSave 85%$0.1000in$0.4000out
- glm-4-6Save 49%$0.1741in$0.7087out
Choose. Constrain. Call.
UsePod sits between your agent and every provider that can serve its request. You describe what an acceptable route looks like — the router does the rest, on every call.
Pick a model
Browse open-weight models served by marketplace operators, alongside commercial providers. Compare price, context, and live availability before you commit.
{
"model": "deepseek-v4-flash",
"messages": [
{ "role": "user", "content": "Hello" }
]
}Set the policy every route must meet
Set price ceilings and choose whether routing may fall back to centralized providers. The router enforces these headers on every request.
X-Pod-Max-Price-Input: 400000
X-Pod-Max-Price-Output: 600000
X-Pod-Routing-Mode: marketplace-onlySend the request you already send
Same OpenAI or Anthropic request shape, same streaming, same tool calls. The response identifies the route and provider that served it.
X-Pod-Route: marketplace
X-Pod-Provider-Id: <provider-uuid>
X-Balance-Cost-Microunits: 114A market, not a walled garden.
Open-weight models close the gap with frontier releases every quarter. When any provider can serve the same model, price converges on marginal cost — UsePod is where that convergence happens.
Real price discovery
Providers compete on identical, content-addressed model weights. You see every route's price before it serves a token, and pay the marginal cost of compute, not a brand premium or subscription tier.
Privacy as a routing constraintTEE soon
Choose which route types may serve each request: marketplace-only, no-retention, region-pinned. Hardware-attested private inference via TEE enclaves is shipping next on the roadmap.
One compatible API
OpenAI and Anthropic request shapes, streaming, and tool calls — served from a single endpoint. Migrating an existing agent is a base-URL swap, not a rewrite.
Fallback that is never silent
If no marketplace route meets your policy, UsePod can fall back to commercial providers you've allowed and says so in the response metadata. You always know who served the request and why.
What's live, and what's next.
Trust claims should be checkable. Here is exactly where the stack stands: routing controls and cost transparency are live today; the hardware-enforced privacy layer is in active development and marked accordingly.
The goal is inference where privacy is proven by attestation, not promised by policy.
Two lines to migrate.
UsePod speaks the API your agent already speaks. Point your existing SDK at the endpoint and every request becomes a marketplace request — routing, fallback, and payment handled underneath.
- OpenAI and Anthropic compatible — same shapes, streaming, and tool calls
- Routing policy per request or per key: price, latency, route types, fallback
- Pay by card or USDC balance — or fully agent-native, per request, with x402
- LangChain, LlamaIndex, and CrewAI integrations
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.usepod.ai/v1", // line 1
apiKey: process.env.USEPOD_API_KEY, // line 2
});
const completion = await client.chat.completions.create({
model: "deepseek-v3.2",
messages,
usepod: { // optional policy
max_input_price: 0.25,
routes: ["marketplace", "commercial"],
fallback: "explicit",
},
});// An agent with a funded wallet pays per request:
// 1. Request hits a 402 paywall with a payment spec
// 2. SDK signs a USDC micro-payment from the wallet
// 3. Retries with the X-Payment header — <500ms round trip
import { PodClient } from "@usepod/sdk";
const client = new PodClient({ wallet: agentWallet });Controls your procurement team will ask about.
Agents spend autonomously; organizations still need governance. UsePod gives both sides what they need — deterministic limits for the machines, visibility and predictable billing for the humans.
Programmatic spending limits
Per-key, per-team, and per-agent budgets enforced deterministically — no inference required to decide whether to spend.
Route governance
Org-wide policies for allowed providers, route types, regions, and retention — applied before any developer's request leaves the router.
Usage & audit exports
Per-request receipts roll up into exportable usage records: model, route, provider, and cost for every call your fleet makes.
Billing that fits procurement
Card and invoiced billing for organizations, USDC balances for treasuries, x402 for fully autonomous agents — on one account.
Turn idle capacity into a route.
Have GPUs or unused API quota? Serve the marketplace at your price.
Any operator can list open-weight models at a price they set. Demand routes to you whenever you're the best match for a request's policy. Payouts settle in USDC per request served — operators keep 90% of every inference fee.
Today, trust is contractual and monitored. With TEE attestation, it becomes cryptographic: clients verify your node runs unmodified code, and reputation stops mattering.
Become a providerThe future of software is agent-first.
The infrastructure that serves it must be built to a different standard: private by hardware, sovereign by design, machine-speed by necessity. That is where UsePod is headed — a permissionless inference protocol where payment, privacy, and verification are properties of the network, not promises of a company.
“A privacy policy is a legal instrument, enforced after the fact — if at all. There must be a separation of mind and state: an agent's reasoning isolated not by policy, but by mathematics and hardware.”