The trust problem with AI decisions
Here's the uncomfortable truth about AI-assisted decision-making: when you ask a model to argue both sides of a question, you still have to trust that the arguments are grounded in reality.
V2 gave you structured multi-agent debate, action pipelines, and outcome tracking. But the arguments themselves? They came from the model's training data. No sources. No verification. No way to check if the AI agent citing "industry benchmarks" was citing anything real.
V3 fixes this. Every argument now comes with receipts.
What V3 ships
1. Evidence Engine
This is the headline feature. During every debate, AI agents now perform real-time web research to find supporting evidence for their arguments.
Here's what that looks like in practice:
- Each agent searches the web for relevant sources before constructing arguments
- Claims are backed by inline citations — click any
[1]marker to see the source, relevance score, and a snippet - A fact-checker agent runs after each round, cross-referencing claims against retrieved evidence
- The final verdict includes a citation reference list with every source used
This isn't retrieval-augmented generation bolted on as an afterthought. Evidence retrieval is woven into the debate pipeline — agents can challenge each other's sources, and the fact-checker flags unsupported claims.
The result: you can actually verify why the AI reached its conclusion.
2. Knowledge Connectors
Your decisions don't happen in a vacuum. They happen in the context of your team's existing knowledge — project docs, issue trackers, internal wikis, shared drives.
V3 introduces Knowledge Sources that let you connect your tools directly to the debate:
- Linear — pull in project context, issue details, and team priorities
- GitHub — reference repositories, PRs, and technical documentation
- Confluence & Jira — connect your team's knowledge base and project tracking
- Google Drive — bring in shared documents, spreadsheets, and presentations
- File uploads — drop in PDFs, DOCX, or any document for the agents to reference
When knowledge sources are connected, debate agents can reference your actual data instead of making generic arguments. "Should we refactor the auth module?" becomes a much better question when the agents can see your actual codebase and open issues.
3. Decision Frameworks
Not every question needs the same analytical approach. A hiring decision and a technical architecture decision require fundamentally different thinking structures.
V3 adds structured decision frameworks that guide the debate:
- SWOT Analysis — strengths, weaknesses, opportunities, threats
- Cost-Benefit Analysis — quantified trade-offs with weighted scoring
- Risk Assessment — probability × impact matrices
- Stakeholder Analysis — map affected parties and their interests
- Custom frameworks — define your own analysis structure
Select a framework before starting the debate, and the AI agents will structure their arguments around it. The verdict synthesizes findings into the framework's format, giving you a structured output you can share with stakeholders.
4. Platform Distribution
V3 is no longer just a web app. We've opened up the platform:
- TypeScript SDK — integrate verdicts into your own applications
- CLI tool — run debates from your terminal or CI/CD pipelines
- REST API — full programmatic access to debates, verdicts, and evidence
- Browser extension — highlight any text on the web, get a verdict (coming soon)
- Raycast extension — quick verdicts without leaving your workflow (coming soon)
What this means for your workflow
The core loop is the same: ask a hard question, get a multi-perspective analysis, receive a verdict with confidence scoring.
But now:
- You can verify claims — every argument links to its sources
- Context is real — agents reference your actual project data, not generic assumptions
- Analysis is structured — frameworks ensure the right questions get asked
- Access is everywhere — use the platform however fits your workflow
V3 is live now for all users. Evidence Engine is available on all plans. Knowledge Connectors and Decision Frameworks are available on Pro and Teams.
What's next
We're working on outcome prediction — using your historical decision data to estimate the likelihood of different outcomes before you commit. More on that soon.
If you've been on the fence about trying AskVerdict AI, V3 is the version to start with. The Evidence Engine alone changes how much you can trust AI-assisted analysis.
Try it now — the first 500 founding members get 3 months of free BYOK Pro access.