Retrieval assistant grounded in this site · answers cite their sources
Paste a sales email, a support ticket or an invoice and run it through the kind of workflow I build for clients: parse, classify intent with confidence scores, extract the fields, route it and take the actions. It runs entirely in your browser, and you can edit the input.
A model scores every possible intent and turns the scores into probabilities. Low confidence goes to a human instead of guessing.
Names, amounts, dates and IDs are pulled into structured fields, so other systems can use them.
Plain business rules decide what happens next: limits, priorities and approvals stay in your control.
The agent does the work across your tools (CRM, Slack, email, accounting) and logs every step.
This demo runs on your text in your browser. Production versions use Claude or GPT with n8n or Make, connected to your real tools, with guardrails and human review where it matters.
Lead a software project from kickoff to maintenance. Every decision moves budget, schedule, quality and stakeholder trust.
OpenAssemble a structured Claude prompt with roles, context, examples and constraints, scored live against best practices.
OpenWatch retrieval-augmented generation work step by step: chunks, similarity scores, the augmented prompt and a grounded answer.
OpenScore features by Reach, Impact, Confidence and Effort, tag them MoSCoW, and get a ranked roadmap you can export.
Open