Where your code and key go
Use the model you choose with your own API key. Your files and key pass through our server in memory to that provider. Bounty Operator stores no code, prompts, keys or results.
No API key? Code security, Solidity and draft-report reviews also work in ChatGPT or Claude, with no daily limit.
- Key
- Yours. Used for the one request it belongs to.
- Files
- Sent through our server to the provider you choose. Not stored by Bounty Operator.
- Method
- Added on our server. Open source for the three core profiles.
- Providers
- 13, each called at one fixed endpoint
- Chat plan
- Copy the prompt into ChatGPT or Claude, paste the answer back (3 review types)
- Coding agent
- Claude Code, Codex or Cursor over MCP
- Record
- A packet with a SHA-256 manifest of every file
Where your code goes
A hosted review has three stops.
-
Your browser
- Reads the files you choose.
- Runs the privacy check on every file and on your instructions.
- Shows the request before it leaves the tab.
- Hashes each file with SHA-256 and builds the review packet you download.
-
bountyoperator.com
- Receives your files, your request and your API key.
- Checks your allowance and runs the privacy check again.
- Adds the review method of the profile you picked and sends the request, with your key, to one fixed endpoint of the provider you chose.
- Streams the review back. Writes none of it to the database or to logs.
-
Your provider
- Receives your files, your request and our review instructions.
- Runs the model you picked.
- Bills the usage to your account.
A GitHub import goes from your browser straight to GitHub (api.github.com and raw.githubusercontent.com) at a pinned commit. The repository is never uploaded to our server.
Three ways to run a review
All three give the same output format and the same packet.
Your API key
Paste a key from one of 13 providers and run the review on this site. It goes through our server to your provider, and the answer streams into finding cards.
- All 11 review types, the Gauntlet and Panel review
- Free: 1 review a day, any review type
- Operator: unlimited, 4 at once
- The key is used for that request only
Your ChatGPT or Claude chat
For 3 review types you get the full prompt as one document. Paste it into your chat app or a local model, then paste the answer back.
- Code security review, Solidity review, challenge a draft report
- No API key, no account and no daily limit
- Your files stay in your browser until you paste the prompt
- The pasted answer gives you the same findings and download
MCP, your coding agent
Connect an agent to the MCP server. prepare_review builds the request of a core profile for the agent’s own model. run_review runs any profile as a hosted review, and is the one way to run a hosted profile from an agent.
list_profiles,prepare_reviewandbuild_packetneed no accountrun_reviewneeds a connection token from your account- Your provider key travels in the
X-Provider-Keyheader, never as a tool argument
Who holds the key
You do.
You create the key in your provider’s console and paste it in when you run a review. It travels with that request to our server, goes on to the provider’s endpoint, and is gone when the request ends.
Each provider is called at one fixed endpoint. A redirect is refused. An error message from the provider is shown to you with the key removed.
Operator is US$10 per month for unlimited reviews, including US$2 in Bounty Operator API credit per paid month. Other model providers bill usage to the account that owns the key.
The privacy check
It runs before anything is sent, in your browser and again on the server.
- Private keys, API keys, GitHub and Slack tokens, wallet keys and seed phrases block the request.
- A file that normally holds credentials blocks the request.
- An email address or a public IP address asks you to confirm before sending.
- A finding names the file and the line. It never repeats the secret.
Up to 50 files, 120 KB per file, 240 KB and 20,000 lines in total.
What is stored
The full list is on the privacy page, and the security page shows how it is protected.
Never stored
- Your files and your code
- Your prompts and review instructions
- Your provider API key and any GitHub token
- The review the model returns
Stored
- A random account identifier, passkey public keys and a hashed recovery code.
- Hashed session tokens and hashed connection tokens, with their labels and last-use times.
- Review activity for up to seven days: status, profile, web or MCP, timestamps. It enforces the daily allowance.
- Billing references from Stripe: customer and subscription ids, status, paid-until and the receipt email.
- Daily totals per page and per event, with no account identifier.
Providers supported
One key from any of these runs every review type. The default model is preselected and the others are suggestions in the model list.
| Provider | Default model | Also suggested | Key starts with | Account and API key |
|---|---|---|---|---|
| OpenRouter | Default modelanthropic/claude-sonnet-5.5 | Also suggestedopenai/gpt-6.1-soldeepseek/deepseek-v4.1-flashanthropic/claude-opus-5.5openai/gpt-6-astragoogle/gemini-3.8-flashx-ai/grok-4.7z-ai/glm-5.3-flashxiaomi/mimo-v2.6-promistralai/mistral-large-4-0moonshotai/kimi-k3openai/gpt-6-luna | Key starts withsk-or- | Account and API keyopenrouter.ai |
| Anthropic | Default modelclaude-opus-5-5 | Also suggestedclaude-sonnet-5-5claude-fable-5-1claude-haiku-4-5 | Key starts withsk-ant- | Account and API keyconsole.anthropic.com |
| OpenAI | Default modelgpt-6.1-sol | Also suggestedgpt-6-astragpt-6-luna | Key starts withsk- | Account and API keyplatform.openai.com |
| Google Gemini | Default modelgemini-3.8-flash | Also suggestedgemini-3.1-pro-previewgemini-3.5-flash-lite | Key starts withAIza | Account and API keyaistudio.google.com |
| xAI | Default modelgrok-4.7 | Also suggestedgrok-4.6grok-4.3 | Key starts withxai- | Account and API keyconsole.x.ai |
| DeepSeek | Default modeldeepseek-v4-pro | Also suggesteddeepseek-flash | Key starts withsk- | Account and API keyplatform.deepseek.com |
| Mistral | Default modelmistral-medium-latest | Also suggestedmistral-large-2512mistral-small-latest | Key starts withNo fixed prefix | Account and API keyconsole.mistral.ai |
| Groq | Default modelopenai/gpt-oss-120b | Also suggestedqwen/qwen3.8-27bopenai/gpt-oss-20b | Key starts withgsk_ | Account and API keyconsole.groq.com |
| Novita AI | Default modeldeepseek/deepseek-v4.1-flash | Also suggestedzai-org/glm-5.3-flash | Key starts withNo fixed prefix | Account and API key Create a Novita AI account (affiliate link). We may earn a commission if you sign up and pay through this link. |
| Nebius Token Factory | Default modeldeepseek-ai/DeepSeek-V4.1-Flash | Also suggestedzai-org/GLM-5.3-Flashmoonshotai/Kimi-K3 | Key starts withNo fixed prefix | Account and API keytokenfactory.nebius.com |
| Infron | Default modelz-ai/glm-5.3-flash-cybersecurity | Also suggestedz-ai/glm-5.3-abliteratedz-ai/glm-5.3-flash-abliterateddeepseek/deepseek-v4-flash-vision-exp-cybersecurityqwen/qwen3.8-27b-cybersecuritygoogle/gemma-4-26b-a4b-it-cybersecuritynvidia/nemotron-3.5-content-safety | Key starts withNo fixed prefix | Account and API keyinfron.ai |
| OrcaRouter | Default modelorca/orcacyber-zero-1.5 | Also suggestedorca/orcacyber-zero-1.0openai/gpt-6.1-solanthropic/claude-sonnet-5.5deepseek/deepseek-v4.1-flashz-ai/glm-5.3-flashqwen/qwen3.8-27b | Key starts withsk-orca- | Account and API keywww.orcarouter.ai |
| Bounty Operator | Default modelcyber-qwen-27b-awq | Also suggestedorca-cyber-zero-1.5glm-5.3-flash-cyberdeepseek-v4-flash-cyberqwen3.8-27b-cybergemma-4-26b-cyberglm-5.3-cyberglm-5.3-ablitglm-5.3-flash-ablitcyberstrike-offsec-35bqwen-coder390-27b-ablitkimi-k3-cyber | Key starts withbo_api_ | Account and API keybountyoperator.com |
Affiliate commission does not determine defaults or benchmark rankings. Novita signup is optional; existing users can go straight to key management. Check Novita’s pricing and credit before a review. Its usage is billed separately, and Bounty Operator’s account and hosted-review limits still apply. Setup and cost details.
Questions
Can I use my ChatGPT or Claude subscription?
Yes, for 3 review types: Code security review, Solidity review and Challenge a draft report. Copy the prompt into your chat, then paste the answer back. You get the same findings and download as with an API key. The other review types need an API key, which providers bill separately from a chat subscription. Claude Max and Team plans include a monthly API credit: see the next answer.
I am on Claude Max or Team. Do I have to buy API credit as well?
Since October 2026 those plans include monthly credits for the Claude API: US$100 on Max 5x, US$200 on Max 20x, and up to US$500 pooled on Team. Link a Claude Console organization in the billing settings on claude.ai, create an API key in it, and reviews on that key draw on those credits first. New subscribers can claim them after seven days on the plan. Free, Pro and Enterprise plans are not eligible. The terms are in Anthropic’s article on the credits.
If Claude blocks a review, see When the model refuses.
Does Bounty Operator see my API key?
The key passes through our server in memory for the one request it belongs to and goes to your provider’s endpoint. It is not written to the database or to logs.
What does your server add to my request?
The review instructions: the output format every review shares, and the method of the profile you picked. The three core profiles carry their method in the open, so their preview is the whole prompt. Every other profile is hosted: the preview shows the request that leaves your tab, with your focus, your context and the files, and the server adds the method before it goes to your provider.
Who bills the model usage, and what does the US$10 pay for?
The US$10 monthly plan pays for unlimited reviews, the Gauntlet, Panel review and 4 reviews running at once. It includes US$2 in Bounty Operator API credit per paid month. Other model providers bill usage to your own key.
What does the provider keep?
The provider processes the request under its own terms and retention settings. Read them for the key you use, and run a review only on material you are allowed to share with that provider.
Can I run it on a free model?
On OpenRouter you can: type the model’s identifier in the Model field. Think twice with a finding nobody has reported. Some free models are served only by hosts that may train on what they are sent. OpenRouter lets an account rule those hosts out in its privacy settings, and a request to such a model then fails: the workbench says the account’s settings left no host for it. Leave that setting as it is and pick a model that is not free. A free model also runs on a request allowance its host shares among all of OpenRouter’s users, and a review fails while that allowance is used up.
Which key does Panel review use?
One OpenRouter, Infron, OrcaRouter or Bounty Operator key covers a whole panel: it reaches models from several labs, so two to four different models review the same files. A panel model on another provider takes that provider’s own key.
Run it on the model you already pay for
Paste a key, or export the prompt of a core profile to your chat app.
Free: 1 review a day. No card.