Early access · Fix agents live
Your agent is stuck.
Call a specialist.
One key for your coding tools, plus fix agents that turn an error into a cause and a patch. Pay per call from one balance in US dollars, with a quote up front and a receipt after.
One key for your coding tools. A specialist when they get stuck. A receipt for every charge.
01What you get
A gateway for your coding tools, and specialists behind it.
The gateway is live today. Fix agents run on top of it, so routing, metering and receipts are built once and shared.
- Live
Fix agents
When your own agent keeps patching the same bug, hand the error to a specialist. It reads the log and the files involved and returns the cause, a patch and a way to check it.
- Quick Fix and Build Doctor, live now
- Returns a patch; never runs commands on your machine
- Economy, Balanced or Strongest model per call
- Live
One key for every model
Qwen3.8-27B, hosted by Inferiq, behind one OpenAI-compatible endpoint. Three aliases set how hard it thinks: inferiq/fast, inferiq/balanced and inferiq/best. A failed attempt is retried once.
- /v1/chat/completions with streaming and tool calls
- Point Cline, OpenCode or an OpenAI SDK at it
- Anthropic-format endpoint for Claude CodeLive
- Live
Pay per call from one balance
One balance in US dollars. Every request and every fix is deducted from it. No subscription, no monthly minimum, no card on file.
- Deposit ETH or USDG on Robinhood Chain, or ETH or USDC on Base
- Balance credited by hand during early accessLive
- Live on mainnet; ETH converted at the rate when credited
- Live
A receipt for every charge
Model, tokens, cost and balance, fetchable by ID.
- Live
A hard maximum per fix
You approve the maximum first, and a run never charges more than it.
- Live
Code not stored after a run
Your files are kept only until the run ends. Results are deleted after 24 hours; receipts keep metadata.
- Live
Free keys, one per tool
Limit each key to the models you want it to spend on. Stored as a hash, shown once.
02Fix agents
A specialist for each kind of failure.
Each agent follows its own procedure, not just a different prompt. Build Doctor reads your manifest and lockfile and checks the installed versions before it suggests a patch.
Quick Fix
LiveOne error and its stack trace in, the cause and a small patch out.
- Takes
- Error message, stack trace, the files it points at
- Returns
- Cause, patch, how to check it
- Economy
- $0.02–$0.035
- Balanced
- $0.035–$0.10
- Strongest
- $0.10–$0.28
Build Doctor
LiveInstall, build and deploy failures: npm, pnpm, Next.js, Vite, Docker, Vercel.
- Takes
- Build log, package.json, lockfile, config files
- Returns
- Cause, patch, the command that should pass
- Economy
- $0.065–$0.15
- Balanced
- $0.15–$0.45
- Strongest
- $0.36–$1.50
Type Fixer
Next · Phase 5A wall of TypeScript errors, fixed at the root instead of one by one.
- Takes
- tsc output, the files involved
- Returns
- Grouped causes, one patch
- Economy
- $0.035–$0.10
- Balanced
- $0.10–$0.35
- Strongest
- $0.28–$0.70
Dependency Doctor
Next · Phase 5Version conflicts, peer dependencies and breaking changes after an upgrade.
- Takes
- Install log, manifest, lockfile
- Returns
- Versions to pin or change, migration notes
- Economy
- $0.065–$0.15
- Balanced
- $0.15–$0.45
- Strongest
- $0.36–$1.05
Test Fixer
Next · Phase 5A failing test: is the code wrong, or the test?
- Takes
- Test output, the test, the code under test
- Returns
- Verdict, patch to the right side
- Economy
- $0.07–$0.18
- Balanced
- $0.18–$0.70
- Strongest
- $0.35–$1.05
Deep Debug
LaterA bug across many files that your own agent keeps going around in circles on.
- Takes
- Symptom, what was tried, the repo files involved
- Returns
- Root cause, multi-file patch, verification plan
- Economy
- $0.35–$0.90
- Balanced
- $0.70–$2.10
- Strongest
- $1.40–$3.50
Prices are illustrative ranges per call, set before real runs give us cost data. Every run gets its own quote with a hard maximum first, and you approve it before anything is charged.
How pricing works03Try it
Paste an error. Approve the price. Get the fix.
A simulated run, so you can see the whole flow first. No sign-in, nothing is sent or charged. Real fixes run from the Fix page in the console.
fix run · simulated
$ inferiq fix --agent build-doctor --tier balanced
Pick an agent and a tier, then get a quote. The example error is already filled in.
04Models
Three tiers. One model, three depths of thinking.
Renaming a variable doesn’t need deep thinking. Every alias runs on Qwen3.8-27B, hosted by Inferiq: fast thinks briefly, balanced thinks more, best thinks hardest. A failed attempt is retried once before anything reaches your tool.
- Economy$0.41 in · $2.02 out /1M
Clear, common errors. Cheapest.
inferiq/fastqwen3.8-27b · light reasoning · retried once
- Balanceddefault$0.41 in · $2.02 out /1M
The default. Most bugs land here.
inferiq/balancedqwen3.8-27b · medium reasoning · retried once
- Strongest$0.41 in · $2.02 out /1M
Tangled bugs across many files.
inferiq/bestqwen3.8-27b · full reasoning · retried once
curl https://api.inferiq.run/v1/chat/completions \
-H "Authorization: Bearer $INFERIQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "inferiq/balanced",
"stream": true,
"messages": [
{"role": "user", "content": "Why does this test fail? Here is the output."}
]
}'
# Response headers
# x-inferiq-model: qwen/qwen3.8-27b
# x-inferiq-receipt-id: rcpt_…Prices are per 1M tokens and come from the gateway’s model catalog. Deeper reasoning writes more tokens, so best costs more per request than fast. The model sleeps when idle, so the first request after a quiet spell can take a few minutes.
05How it works
Deposit once. Every call draws from the same balance.
Inferiq sits between your coding tool and the model providers. Every request is checked against your balance, routed, metered and written down.
- 01
Your coding tool
Cline, OpenCode, Cursor or your own script. It only knows a base URL and an Inferiq key.
- 02
Inferiq
Checks the key, the models it may use and the balance, then picks a model or runs a fix agent.
- Key, model & balance check
- Quote & hold (fixes)
- Alias router & fallback
- Metering & receipt
- 03
The model
Qwen3.8-27B on Inferiq’s own GPUs, so you don’t need a provider account. It wakes up when a request arrives and sleeps when nobody is using it.
- 04
Back to you
The response or the patch, plus the model that answered and a receipt ID.
From deposit to receipt.
No subscription and no provider accounts. Top up when the balance runs low; nothing renews on its own.
Read the quickstart- 1
Add balance
LiveDeposit ETH or USDG on Robinhood Chain, or ETH or USDC on Base on mainnet. ETH is converted to US dollars when it’s credited. During early access, Inferiq can also credit balance by hand.
ETH · USDG · USDC → USD balance - 2
Connect your tool
Create a free key for each tool and pick the models it may use. Paste the base URL and the key into any OpenAI-compatible tool, or add the MCP server to call fix agents from your editor.
base URL https://api.inferiq.run/v1 - 3
Call
Chat requests run straight away. A fix comes back as a quote with a hard maximum, and runs only after you approve it.
quote $0.28 · max $0.45 → approve - 4
Get a receipt
The final cost is deducted and any unused hold goes back to your balance. The receipt shows both.
x-inferiq-receipt-id: rcpt_…
06Receipts
Every charge has a receipt.
A coding agent can burn through millions of tokens in one session without you noticing. Every gateway request and every fix writes a receipt: which model answered, how many tokens, what it cost and what your balance was before and after.
- Before
- A quote with an estimate and a hard maximum. The maximum is held from your balance.
- After
- The real cost is charged and the rest of the hold is released, in the same entry.
- Later
- Fetch any receipt with GET /v1/receipts/{id}, or browse them in the console.
Every request and every fix writes a receipt today, with quotes and holds on fixes.
Usage and receipts in the docsrcpt_fx8c21a4
16 Oct 2026 · 14:23:00 UTC
- kind
- fix.run
- agent
- Build Doctor
- tier
- Balanced
- model
- qwen/qwen3.8-27b
- tokens in
- 18,420
- tokens out
- 2,140
- duration
- 24.8 s
- quote estimate
- $0.2800
- quote max
- $0.4500
- held
- $0.4500
- released
- $0.2008
- balance before
- $41.65
- balance after
- $41.40
- CHARGED
- $0.2492
Amounts in USD.
07Connect your tool
A base URL and a key. That’s the setup.
Coding tools only need to know where to send requests and which key to send. Paste these in and keep working the way you already do.
A free key for each tool.
Limit each iq_ key to the models that tool should spend on. Revoke one without touching the rest.
A receipt ID on every response.
The x-inferiq-receipt-id header links any reply back to its cost.
Clear about limits.
Where a tool can’t use a custom base URL for a feature, we say so.
Base URL https://api.inferiq.run/v1
API Key iq_…
Model ID inferiq/balancedUse inferiq/fast for Plan mode and inferiq/balanced for Act mode if you want to save on planning turns.
08Roadmap
The gateway has shipped. Fix agents are next.
Each phase ends when it can be tried for real, not on a date. Anything on this page that isn’t done is labeled.
- Phase 1Done
Core gateway
One key for your coding tools, a receipt per request.
- Phase 2Done
First fix agents
Quick Fix and Build Doctor are live and priced before they run; on 60 real errors they returned a correct patch 30/30 and 27/30 times.
- Phase 3Done
Wallet balance
Deposit ETH or a stablecoin once, pay per call.
- Phase 4In progress
Fix from your editor
Call a fix agent without leaving Claude Code or Cursor.
- Phase 5Planned
More agents and a CLI
More specialists, a CLI.