Decisions · BridgeheadOpenAI Decisions × every MCP app
Fast, typed decisions, in every app you work in.
On your own OpenAI key.
Decide · Pictures · Rank · Files · Gate · Recipes · Sets · Many files · Side by side · Checked
OpenAI's Decisions API answers with a typed decision (a yes or no with its probability, a choice, a score), over text and pictures, usually in well under a second. But it's an API: there's no way to reach it from claude.ai, ChatGPT, the phone or voice, and by itself it keeps no record of what it answered or how often it was right. Decisions · Bridge is a connector you add once: sign in, your OpenAI key goes into Bridgehead's vault, and Decisions' yes/no checks, choices and scores are tools in every app you work in, with ranking, questions over a picture or a file, a gate for agents, ready-made recipes, saved question sets and many files at once on top. Every answer says how long it took and what it cost, and is kept with its confidence, to mark right or wrong.
What it can do
- Answer yes or no, with how sure it is.
- Pick one of your options, each with its probability.
- Rate on a scale you describe, with labels for its points.
- Several questions over one text in a single call.
- Set a threshold: below it, the answer comes back as unsure instead of a guess.
- Every answer says how long it took, in milliseconds, and what it cost.
- Up to 50 files at once, from a list or a selection in your gallery.
- A pass over your gallery's pictures or documents, or a list of up to 500 items.
- Before it starts: how many items, and about what it will cost. Nothing is spent until you say go.
- A cap in dollars: it stops before the item that could take it past.
- Answers are kept on the items; ask for only the ones that answered yes, or a category, or a score.
- With a saved question set, items already answered aren't paid for twice.
- Over a text, a picture, a PDF or a text file, with your questions or a saved set.
- Question by question: each one's answer, and whether they agree.
- Each engine's time, measured from the same place at the same moment, and what Decisions cost.
- A running tally of your comparisons so far.
- Image comparison with Jev requires a Gemini key: Jev can't take a picture by itself, so Gemini reads it into text first. Text comparison is unaffected.
Copy. Paste. Connected.
How to connect it
MCP server
The address to paste into each app, or its one-line command. It's the same connector everywhere.
Claude
claude.ai, the desktop app or the phone: Customize → Connectors → Add → Custom connector, then paste this.
https://decisionsbridge.fyi/mcpClaude Code
One line in your terminal.
claude mcp add --transport http decisionsbridge https://decisionsbridge.fyi/mcpChatGPT
In settings, turn on developer mode and add a custom app with this address.
https://decisionsbridge.fyi/mcpCodex
One line in your terminal; it asks you to sign in.
codex mcp add decisionsbridge --url https://decisionsbridge.fyi/mcpAny other MCP app
Cursor, VS Code, Windsurf, Zed, n8n and more: add this address as a remote MCP server and sign in when asked.
https://decisionsbridge.fyi/mcpYour OpenAI key
Sign in when your chat asks, then add an OpenAI API key that can reach the Decisions API (a public beta). You pay OpenAI directly.
Keys it uses: your OpenAI key; your Gemini key, for recordings, video and scans; your Jev key, for the side by side with Jev. A key already in your Bridgehead vault is used as it is: one tap, nothing to paste twice.
Add your OpenAI keyQuestions
What can I do with Decisions · Bridge?
Ask a yes or no question and get the probability; pick one of your options; rate on a scale you describe; or ask several questions over one text in a single call. Rank many candidates against a query, and ask your questions over a picture, a PDF or a text file. Set a threshold, and an answer below it comes back as unsure instead of a guess. Save a set of questions under a name and reuse it on any text or file, and let an agent ask the gate before it runs a command or edits a file: allow, ask or block, with the reasons. Recipes on top: triage a message, screen text before it's used (prompt injection included), sort a text into your own categories, and pick a fast or a powerful model. Up to 50 files can be judged at once in the background, also from a selection in your gallery, and every answer is kept with its confidence: mark answers right or wrong and see how often Decisions is right. Every answer also says how long it took and what it cost.
What does a pass over my gallery cost, and can it run away?
It tells you first, and it can't. Ask for a pass over your pictures or documents, up to 500 items, and it answers with an estimate before anything is spent: how many items, and about how many dollars. A picture costs by its size, a few hundredths of a cent at most, so a few of your own are measured for the estimate. Nothing starts until you say go. It then runs in the background under a cap in dollars, twice the estimate unless you set one, and stops before the item that could take it past. You can see what it has cost so far, stop it, and ask for only the items that answered yes, or a category, or a score. With a saved question set, items it has already answered aren't paid for twice.
Can Decisions look at pictures?
Yes, directly. A picture goes to OpenAI's Decisions API as it is, with your questions beside it, so nothing has to describe it first and no second key is needed. A large picture is made lighter before it's sent, which keeps the answer quick. A PDF with selectable text and a text file are read by Bridgehead and judged as text.
What about recordings, video and scanned PDFs?
The Decisions API takes text and pictures only. For a recording, a video or a scanned PDF, Gemini reads the file into text first, on your own Gemini key, and Decisions answers over that text. The reading is kept for you alone, so asking other questions about the same file doesn't pay for it twice. Without a Gemini key, it tells you where to add one.
How is this different from Jev · Bridge?
It's the same toolkit on a different engine. Jev · Bridge answers with TypeSafe's Jev on your Jev key; Decisions · Bridge answers with OpenAI's Decisions API on your OpenAI key. Decisions takes pictures directly. Two of Jev's tools aren't here yet: pulling named values out of a text, and checking claims against a source. With both keys added, the side by side asks the two the same questions at the same moment and shows what each said, whether they agree and how long each took. For a picture, Jev needs your Gemini key as well: Gemini reads the picture into text for it.
How do I know when to trust its answers?
Every answer comes with its confidence. Tell your chat when one was right or wrong and Bridgehead keeps count: how often Decisions was right for each kind of question and at each level of confidence, and the threshold that keeps it right 9 times in 10. Set that threshold, and anything less sure comes back as unsure instead of a guess.
Is my OpenAI key safe?
It's checked with OpenAI, kept encrypted in Bridgehead's vault and used only to call the Decisions API for you. You pay OpenAI directly; remove the key at any time at decisionsbridge.fyi/key.
Does Bridgehead keep what I ask it to judge?
No. Bridgehead keeps each answer, with its confidence and the model that gave it, so you can mark it right or wrong. It doesn't keep the text or the picture that was judged.