Documentation

What is Vesta?

Vesta helps you choose between options. You say what matters to you, such as price, quality or risk, and how much each thing matters. Vesta scores every option and gives you a ranking you can explain to others.

A project moves through six stages:

Define what matters (criteria)
Say how much each one matters (weights)
Add the options and their values (alternatives)
Order the options the numbers left level (trade-offs)
Read the result and how stable it is
Close the decision and share the report

AI features explain results and point out inconsistencies. They read your project and leave your data as it is.

Quick start

1. Create a project

In your project list, click New Project (the + button on a phone) and pick one of four starts:

Blank: a name and an empty model
Template: a ready-made model, such as vendor evaluation, hiring or tool selection
Import: a Vesta file (.json) exported from another project
Create with AI: connect your own AI assistant, then use the Vesta creation command to discuss and preview a model before creating it. Open the project link your assistant provides.

Each start asks for a short description: what you are deciding and why. Collaborators read it first. The owner edits it later by clicking it in the project header.

2. Add criteria

In the Criteria stage, add each factor that matters. Pick its type: number, rating, labels or yes/no. Add a veto for any hard limit, such as a maximum budget.

3. Set weights

In the Weights stage, drag the criteria into order of importance, or compare them two at a time. You can switch between the two ways at any time.

4. Add alternatives

In the Alternatives stage, add each option and fill in its value for every criterion.

5. Settle the trade-offs

The Trade-offs stage lists the options that tie on a criterion. Put each group in order.

6. Read the result and close it

The Results stage shows the ranking, how each score adds up, and how far a weight must move to change the winner. The adjudicator then closes the decision, which locks the project and completes the report.

Criteria

Criteria are the factors you judge each option on. There are four types.

Quantitative

A number with a unit, such as price in $, weight in kg or response time in ms. You choose whether higher or lower is better. Two options tie when their values are within 3% of the full range.

Ordinal

A rating on a fixed scale, such as 1–5 stars or 1–10, entered with a slider or stars. Higher is better. Options with the same rating tie.

Mapped

Named labels, each worth a score: Small → 2, Medium → 5, Large → 10. Good for company size, support tier or contract type. Mark a label as a veto to rule out every option that carries it.

Boolean

Yes or no. You pick the preferred answer, and optionally the answer that triggers a veto. Good for compliance checks and required features.

Vetoes

A veto rules an option out. A vetoed option scores zero, sits at the bottom of the ranking and carries a "vetoed" label. Its potential score, the score it would have had, stays on screen.

Alternatives

Alternatives are the options you choose between: vendors, candidates, products, plans.

Entering values

Enter a value for each criterion. The input follows the type: a number field, a slider or stars, a dropdown, or a toggle.

Blank cells

An empty cell scores at the bottom of the scale, so missing data lowers a score. Vesta keeps blanks out of the best-to-worst range, so a blank price stays out of the cheapest spot. The report prints a blank as "not recorded".

Properties

Add details to any option, such as a link, a location or a note. They show in the results and the report. Scores ignore them.

The starred alternative

Star an option to test your favourite. The star shifts weight toward the criteria that option is strong on (the Vesta adjustment). Results then shows every score twice, with and without the star, and tells you whether the order changed. Remove the star to remove the adjustment. A starred option stays in the ranking even when vetoed.

Weights & comparison

Vesta works out weights with the Analytic Hierarchy Process (AHP). You compare criteria, and Vesta turns the comparisons into percentages. There are two ways to compare.

Simplified mode

Drag the criteria into order, most important first. Vesta gives the top of the list the most weight. This is the default for new projects and the fastest way: ten criteria take one pass instead of 45 comparisons. An order is always consistent.

Pairwise mode

Compare criteria two at a time: "Is price more important than delivery speed, and by how much?" The scale runs from equal to extreme in either direction. Use it when the size of a gap matters, for example when your first criterion matters far more than the second.

Weights and the range of values

A weight says how much it is worth to go from your worst option to your best one on that criterion. The weighting screen shows that range beside each criterion. If all prices are within a few pounds of each other, give price a small weight. A criterion where every option is level shows no spread.

Switching modes

Both modes store the same comparisons, so switching keeps your weights, scores and ranking. A pairwise matrix opened in simplified mode shows as the order of its weights. Your weighting changes only when you drag an item or move a slider.

Pairs you skip

Vesta fills in a skipped pair from your other answers. If price beats quality and quality beats speed, it infers price against speed. A criterion you have yet to compare gets the average weight. Before any comparison, every criterion gets the same weight.

When answers conflict

In pairwise mode your answers can go in a circle: price beats speed, speed beats support, support beats price. Vesta names the three judgements involved. Your weights are still calculated, as the best fit to all your answers. You can fix a judgement if the circle was unintended, or keep everything as it is.

With categories, you weight in two steps: first the categories, then the criteria inside each one. A criterion in an important category counts for more.

Trade-offs

Sometimes two options have the same value on a criterion, such as the same price. The Trade-offs stage lists these ties and asks you which option you prefer. Like Weights, it has two modes, and Vesta remembers your choice for each project.

Simplified mode

The default, and the mode where you settle ties. Each criterion with a tie gets a card. Tied options are highlighted: drag them into the order you prefer.

Then say how far apart they are: barely, clearly, decisively, or they really are equal. A settled card tells you how your choice changed the overall ranking.

One order per tie

If three options tie on a criterion, you order all three at once. That one order settles every pair among them.

How much a trade-off counts

A trade-off nudges the tied options apart. The nudge is small, so it reorders the tied options and leaves the rest of the ranking to your values.

Pairwise mode

A read-only, head-to-head view. Pick two options and compare them criterion by criterion. To settle a tie, use the Simplified cards.

AI guidance

The AI panel on the right reads your settled trade-offs and flags orders that clash with each other. It works in both modes.

Results

The Results stage shows the final ranking. Expand an option to see what each criterion added to its score.

How robust is the result?

A robust result keeps the same winner when you change the weights a little. Under the ranking, Vesta shows each criterion that could change the winner, and how much its weight would have to move.

Score by category

With categories, each option also shows a score for each category. For example, an option can score high on Cost and low on Quality.

Vetoed options

An option that breaks a veto rule scores zero. Vesta also shows the score it would have had, for comparison.

Closing a decision

Closing records which option you chose. It turns the project into a record. The Close the decision button is on the Results stage, for the adjudicator and the owner.

What you enter

The option you chose, and your reason in a line or two (up to 250 characters). The report opens with that reason.

You can choose any option, including one lower in the ranking or one with a veto. The record keeps your choice's position and the option the model ranked first. The report states any difference and names the veto.

You can close with blank values or open comparisons. The dialog counts them, and the report records the same counts.

What closing locks

Criteria, options, weights and trade-offs become read-only for everyone, including the owner. The project shows a "decided" badge in your list and in its header.

You can still read the report and the history, export the project, invite people and make the project public to publish the report.

Opening a project whose decision is closed takes you to its report.

Reopening

The owner reopens a decision from Settings and gives a reason. Reopening is owner-only, whatever roles others hold.

The first closing stays on record, marked as reopened, with both reasons. The report lists it under Earlier closings. Closing again adds a new record beside it.

To change a closed decision, reopen it first, then close it again.

Closing and the scores

Closing records your choice and leaves every score as it was. The report shows the same scores you decided on. Note that this is different from starring: starring reweights the model toward the starred option.

The report

The report explains the decision to people outside the project. Open it from the menu next to the project name, above Settings. Before the decision closes, it tells you what is still needed.

Sections

The decision. The option chosen, by whom, when and why.
The result. Every option in order with its score, the chosen one marked.
What the model knew. The blank values and open comparisons at the time of closing.
Robustness. How far a weight had to move to change the winner.
How this was reached. Turns, handovers, days worked, time to decide, and any role left unused.
Participants. Who did what, for readers with access.
Earlier closings. Decisions closed and later reopened, with both reasons.

The PDF

Download as PDF builds the file in your browser and names it after the project. The PDF adds a summary and annexes with every criterion and option, so a reader can check each score.

A blank value prints as "not recorded".

The AI audit annex

When the project has an AI audit, the download asks whether to include it as an annex. The audit must be of the final model: if the model changed after the audit, the dialog explains why it is unavailable.

Who sees the participation record

The totals (turns, handovers, days worked, time to decide, unused roles) are anonymous and go wherever the report goes. The list of who did what is for the owner and the facilitator. Vesta sends it only to them, so a forwarded PDF carries the totals alone.

On a public project, anyone with the link reads the report. The participation section is for project members only. Public readers see the rest of the report.

Public readers also see members as "member 1", "member 2" and so on. The report still shows how many people held each role.

Reading the process summary

The summary reports on roles, and names people only in Participants. A typical line: "The evaluator scored nothing".

Days to decide runs from the day the project was created to the day it closed. Active days counts the days someone worked on it.

Export

Export saves the project to a file. Open the ⋮ menu beside the project name and choose Export. Anyone who can open the project can export it.

Vesta file (.json)

The whole model: criteria, options, values, weights, trade-offs and the starred option. Keep it as a copy, or load it with Import to start a new project. Vesta recalculates the scores on import. People and roles stay with the original project.

XMCDA file (.xml)

XMCDA is a standard format for decision-analysis tools such as diviz. The file holds the options, the criteria and their direction, every value, the weights, and the categories with their weights. Blank cells stay blank.

Vetoes, trade-offs, the starred option and final scores are only included in the Vesta file. In XMCDA a veto means a gap between two options, and in Vesta it means a fixed cut-off. The XMCDA export lets you import your values and weights into another tool.

AI features

AI features explain and question your model. They read your values and weights, and never edit the project. When the data changes, older insights are marked outdated. AI features are turned off once the decision is closed.

Results insights

On the Results stage: why the top option won, which criterion matters most to the result, and what would happen if a weight changed.

Trade-off guidance

On the Trade-offs stage. The AI reads how you settled the ties and flags orders that clash with each other.

Contradiction check

On the Results stage, for projects built with the previous in-app AI builder. It compares the final weights with what you said in the conversation. For example, it flags a criterion you stressed that ended up with a small weight.

Project audit

On the Audit tab. A check of the whole model:

Dissonance patterns: judgements that pull against each other
Weight suggestions: changes that fit your trade-offs better. The report lists them as suggestions.
Omitted variables: criteria you may have missed

Connect AI

Connect an MCP-compatible AI client to read and edit your decision models, manage alternatives, set preferences and evaluate options, including their descriptive properties. Open your profile to find the connection URL and manage connected clients.

https://mcp.vesta-mcda.com/

Add this URL as a remote HTTP MCP server in your client, such as OpenCode or Claude Code.
Follow the client’s authentication prompt and sign in to Vesta.
Select the projects to share. Enable evaluations to save values, “Manage alternatives” to add, rename, reorder or delete options, and “Edit criteria and weights” to change the model and preference comparisons.
Review the complete selection and approve. Ask your client to list your Vesta projects, inspect a model or evaluate a named alternative.

Your existing role still controls what the client can see and edit. Viewers and facilitators can connect projects for read access. You can approve several projects in one connection. Reauthorizing replaces the previous selection; cancellation keeps it. Disconnecting in your profile revokes the whole connection.

Read the AI connection limits before starting a large batch of work.

Create a project through conversation

Enable “Create new projects” when connecting your client. You do not need to select an existing project. Ask the client to help frame your decision, define measurement scales and name the options. It previews the complete model for your review, then creates the approved project in your account. Evaluation cells stay empty and weights start neutral. Set your preferences and evaluate the options afterward. The creating client can immediately read, evaluate, manage alternatives and revise criteria and weights in the new project. Older projects, including those created through AI before alternative management was introduced, keep their existing permissions. Reconnect and enable “Manage alternatives” for each older project where you want the client to add, rename, reorder or delete options.

In New Project → Create with AI, the setup command comes first, with setup instructions alongside it. The creation command stays available as a reminder. The previous in-app builder is available only to finish a saved draft on the same device; existing projects and their private builder transcripts are preserved. This connection uses your assistant's account and does not require an API key in Vesta.

Continue refining a decision

Return to the same project to add or revise criteria, manage alternatives and change preference comparisons. The client previews the changes, resulting weights and any evaluations or trade-off adjustments that would be removed, then applies your approved proposal. Unaffected evidence and preferences stay intact. Current roles and closed-decision restrictions still apply.

Evaluate several alternatives

With evaluation access, your client can save supported criterion values and descriptive properties for up to 20 existing alternatives together. Every row must be valid for the batch to save; if any row fails, none of the changes are saved. Values you do not update stay intact.

Explore sensitivity and what-if scenarios

If your role permits results, ask which criterion weights could change the winner, or compare a temporary set of preference comparisons or veto thresholds. Vesta calculates the ranking, score and veto changes without saving the scenario. Weight sensitivity varies one criterion at a time; it does not prove robustness against every combination of changes. Saving revised preferences or constraints still requires a separately reviewed edit with the appropriate permission.

Codex setup

Use the DCR registration option because Codex's client-metadata host currently refuses requests carrying Cloudflare's Worker header. This applies to Vesta's production server as well as local development. This command adds Vesta and starts authentication:

codex mcp add vesta --url https://mcp.vesta-mcda.com/ --oauth-client-registration dcr

To change the projects or permissions later:

codex mcp login vesta --oauth-client-registration dcr

Claude Code setup

claude mcp add --transport http vesta https://mcp.vesta-mcda.com/

In Claude Code, open /mcp, select Vesta and follow the authorization flow. Allow the assistant to create new projects when Vesta asks.

Vesta creation commands

Claude Code: MCP command

In Claude Code, a connection named vesta exposes /vesta:vesta-create-project. It comes from the connected MCP server; no separate skill installation is needed. If you chose another connection name, use that prefix instead. If it is missing, reconnect the server and check its commands with /.

Codex: skill command

Codex can use the optional $vesta-create-project skill. Connecting MCP does not install it. From a Vesta repository checkout, install the command pack with:

node scripts/install-vesta-commands.mjs --codex

Restart Codex after installation, then use $vesta-create-project and describe your decision when asked.

If a command is unavailable in either client, or you use another assistant, send this prompt: Use Vesta's vesta-create-project workflow.

Vesta also publishes MCP prompts for reading models, alternatives and results, and evaluating alternatives and creating projects and editing criteria or weights. Whether they appear as slash commands depends on your client. Plain-language requests use the same authorized tools.

Key concepts

AHP (Analytic Hierarchy Process): A method that turns comparisons between criteria into weights. Vesta takes the comparisons in simplified mode or pairwise mode.
Simplified mode: Weighting by dragging criteria into order of importance. Ten criteria take one pass. It is the default for new projects and covers categories too.
Pairwise mode: Weighting by comparing criteria two at a time and saying how much more one matters. It records the size of each gap.
Normalisation: Rescaling values to 0–1 so a price and a star rating can be added. The worst value becomes 0 and the best becomes 1. A score of 0.7 means the option sits 70% of the way from the worst to the best option on that criterion.
Veto: A rule that rules an option out. A vetoed option scores 0.
Tie group: The options that tie on one criterion. One order settles every pair in the group.
Offset: The small score change a trade-off order adds. It stays below the gap to the next option and below a fixed ceiling.
Robustness: How far one criterion’s weight must move to put a different option first. Vesta finds it by rerunning the scoring, and shows it under the ranking.
Starred alternative: An option you mark to test it. The Vesta adjustment shifts weight toward its strong criteria, and Results shows every score with and without that adjustment.
Hierarchical AHP: Weighting in two levels, used with categories. A criterion’s final weight is its category weight times its weight inside the category.
Decision record: What closing saves: the chosen option, who chose it, when, why, its position in the ranking, the model’s first choice, and the gaps in the model. Reopening keeps the record and adds to it.
Outdated insights: AI analysis of an earlier version of your project. Vesta marks it as outdated when the data changes.

Collaboration

Invite people with Share and give each person a role.

Roles

Role	Does	Sees
Viewer	Reads the project	Every stage
Evaluator	Adds options and fills in their values	Criteria and Alternatives
Modeller	Defines criteria and categories, and sets weights	Criteria and Weights
Adjudicator	Settles trade-offs, runs AI insights, closes the decision	Every stage
Facilitator	Manages people, reads the process summary	Every stage

Evaluators and modellers are blinded. An evaluator scores options with the weights and ranking hidden, so each score reflects the option. A modeller sets weights with the options hidden, so the weights reflect what matters. Hidden stages stay in the sidebar with a note explaining why.

Blinding keeps numbers off a person's screen. It is a working method, and the data stays readable: scores are calculated in the browser, and an export holds the whole model. Invite people you trust with the results.

A person can hold several roles and sees everything each role allows. The project's creator is its owner. The owner can do everything and needs no role.

Project actions

These actions apply to the project itself. Most live on the Settings page, reached from the gear at the foot of the stage sidebar or from the ⋮ menu next to the project name.

Action	Who	Why
Export	Anyone who can open it	Opening a project already loads all of its data into your browser.
Manage people	Owner, facilitator	This is the facilitator's job. Their own roles stay with the owner.
Undo and redo	Anyone who can edit	Shown to roles whose changes undo can restore.
Edit the description	Owner	The description frames the decision, and every role reads it.
Reopen a decision	Owner	Reopening keeps the original record and asks for a reason.
Read the history	Owner	The history shows every value, including the ones hidden from blinded roles.
Import data	Owner	Import replaces the whole model at once.
Rename	Owner	Everyone finds the project by its name, including in their invitation.
Public or private	Owner	A public project is readable by anyone with the link.
Delete	Owner	Deleting is permanent for everyone on the project.
Leave	Everyone except the owner	An owner hands the project over or deletes it instead.

The owner can hand the project to a facilitator who has opened it at least once. The facilitator becomes the owner, and you keep access if they invite you back. This step is final.

Inviting

You invite by email address, and the invitation applies at once. The project appears in their project list when that person signs in with that address. New Vesta users find it waiting after they sign up.

Type addresses carefully: a mistyped address looks the same as someone who has yet to open the project. If a colleague reports a missing project, check Settings → Access. It shows “Has not opened it yet” beside each person still to arrive.

Leaving

Leave a project from Settings → Danger zone or from your project list. The project stays with its owner, who can invite you again. Shared projects sit outside your own project limit.

Project notifications

The bell beside your avatar shows when a project needs you: alternatives need scoring, started weights need review, a decision is ready for adjudication, or a decision has been closed. Notifications follow your current role and access. Your own actions do not notify you.

Open a notification to go to the relevant step. Mark it as read, or use Mark all read to clear the current unread count across your devices. Reading a notification does not complete the work. Finished tasks remain in your history while your access lasts; Older notifications loads earlier items. In your profile, Project emails has separate switches for collaborative updates and reminders for your own unfinished projects. When delivery is enabled, updates are combined into at most one email every seven days. Read or completed tasks are omitted. Each email has an unsubscribe link.

A solo reminder needs seven full days without meaningful progress and no other members or pending invitations. Each unfinished step gets one reminder; meaningful progress can make a later reminder useful again. Existing projects start a fresh quiet window when this feature first observes them.

Your project list

Your project list has three tabs. Open holds everything in progress, yours and the ones shared with you. A project moves to Decided once its decision is closed, and back to Open if someone reopens it. Each project says whether you own it or which roles you hold on it.

Finished with a project for now? Archive it from its menu and it moves to the Archived tab. Everyone else on the project still sees it just as before.

Changed your mind? Unarchive it from the same menu, or from the banner at the top of the project. One thing to keep in mind: a project you own still counts toward your project limit after you archive it. If you need the room for a new one, delete it instead.

Project settings

Settings holds the actions on the project itself. Open it from the gear at the foot of the stage sidebar, or from the ⋮ menu next to the project name.

You see the sections your role allows, so the page differs from person to person. The table under Collaboration lists who can do what.

Project

Rename the project and set it public or private. A public project is readable by anyone with the link, signed in or out. A private project is for the people you invite. Owner only.

Access

Invite people by email, give and remove roles, and see who has arrived. “Has not opened it yet” means the person has yet to sign in with that address, or the address has a typo. For the owner and the facilitator.

The owner also hands over ownership here, to a facilitator who has opened the project at least once.

History

A link to the change log. Owner only. See History.

Decision

Appears while a decision is closed, for the owner. Reopen the decision here. The closing stays on record.

Danger zone

Import data replaces the whole model from a Vesta file. Owner only.
Leave this project removes your access at once. The owner can invite you back. For everyone except the owner.
Delete this project removes it for everyone, permanently. Owner only. The change log survives it.

History

Vesta records every change to a project: what changed, who changed it, and the value before. The owner reads it at Settings → History. Vesta keeps the log after a project is deleted.

What an entry says

Each entry names the action, such as “Set the weights”, then each value that moved, from and to. A value change reads Option Suburban Family · Price · value 560000000 → 560000. A weighting reads Practical moderately more important.

Some changes cause others. Deleting a criterion also removes its values and comparisons. Those follow-on changes are listed and marked knock-on.

An undo appears as its own entry, after the change it reverses.

Why the owner

The log shows every value before and after each change. That includes the weights hidden from evaluators and the options hidden from modellers. So the log is for the owner alone.

Limits

Connected AI clients also have request and editing limits.

Vesta is free to use. Your project limit counts the projects you create. Projects shared with you are extra.

Per account

Limit	Value
Projects you create	3

Per project

Limit	Value
Categories	10
Criteria	20
Alternatives	20
Collaborators besides the owner	20
Property keys per alternative	10
Name length	4 to 30 characters
Project description	250 characters
Other descriptions and notes	160 characters
Decision rationale	250 characters

AI features

AI features use your own OpenAI API key. Your browser calls OpenAI directly, so your key and conversations stay between you and OpenAI. The key is stored encrypted on your device. OpenAI bills your own account, and your usage is yours to set.

Using a VPN?

A VPN, proxy or content blocker can stop the call to OpenAI. If an AI feature fails to connect, turn the VPN off for this site.