Envoy keeps the deal at the center: a board of deals across pipelines, each with its health, risk flags, and next steps, and a forecast around them. It is not open to customers yet.
Envoy treats the opportunity as the main object. Each group below is something the code does today. Deal health, risk flags, and next best actions are fixed rules you can read, not a model, and the "AI forecast" is a placeholder; both are explained on the scoring page.
A pipeline is an ordered list of stages, each with a win probability. The demo workspace has three: Fleet Sales, Renewals, and Software and Services. Each column on the board shows how many deals it holds, their total value, and its probability.
Dropping a deal on another column changes its stage, and so does the Stage menu in the deal drawer. A stage can carry entry and exit criteria: a field that must be filled in, a contact role that must exist, a minimum amount, or an activity that must have happened. A hard criterion stops the move until someone types a reason, and a soft one only warns.
Click a card to open its drawer. Overview holds the name, amount, close date, tags, and the health panel. Activity is a timeline where you log notes and see emails, calls, and meetings. History lists what changed on the deal.
Accounts list name, domain, industry, tier, and ARR, with search and archive. Tasks is a plain to-do list with priorities and an open, done, or all filter.
Six signals add up: how likely the stage is to close, how recently there was activity, how much activity there has been, whether the close date is realistic, whether there is an amount, and whether the deal has an owner and a primary contact. It shows as a colored chip on every card.
Ghosted (no activity for 14 days), stage rot (in a stage past its limit), a slipped close date, and no primary contact. A flag that stops being true is cleared, and the cleared flag is kept with its time.
Up to five suggestions per deal, picked by rules: add a primary contact, send a first touch, re-engage after a week of quiet, capture an amount, set a close date, or follow up on a proposal within 48 hours.
Background jobs refresh risk flags hourly, next best actions every four hours, and health scores every six hours, so the board stays current without anyone pressing a button.
For open deals closing in a window, weighted is each amount times its stage probability and best case is the full amount. Commit is the sum of what reps commit. A history table keeps earlier windows. The page computes the current calendar month.
The Reports page runs saved report specs against your deals. Three standard ones come with a workspace: pipeline by stage, win count by rep, and total open pipeline.
Enter a rep and see the year-to-date total earned and each accrual with its rate, amount, and status. Plans with attainment tiers, quotas, clawbacks, and payouts exist in the API.
Settings holds multi-step outreach sequences (send, wait, branch on a reply or an open) that stop when someone replies, and territories whose rules match a lead to reps who each have a capacity.
Pipelines, deals, accounts, activities, tasks, forecasts, quotes, sequences, territories, commissions, and reports are all available as tools an AI assistant can call, over the same REST API the web app uses.
Workers recompute scores and flags, run due sequence steps every 15 minutes, and refresh forecasts hourly, all from a beat schedule on Valkey.
The web app follows your theme and is built on the Renkara design system. Its accessibility results, including what fails, are on the accessibility page.
Real captures of the running tool, with fictional StrataForge Robotics data. They follow your theme.
In Settings, set the stages, their probabilities, and the criteria a deal must meet to enter or leave each one.
Create deals against accounts, log notes and calls, and drag each card along as the deal moves.
Check the health score, the risk flags and their reasons, and the next best actions before the next call.
Compute a window's forecast, run the saved reports, and look up a rep's commission statement.
Envoy's web app has a skip link, a labeled navigation landmark, named dialogs, and labeled form fields, and we ran axe-core on every page and the deal drawer at 375 and 1440 pixels in light and dark. The first run found 25 serious color-contrast problems; we fixed them, and the rerun found no serious or critical violation. A deal moves between stages by dragging on the board or from the Stage menu in the deal drawer, which a keyboard reaches, and the dialogs do not manage focus, which we list below instead of hiding. We do not claim a WCAG level.
Partly supported
A skip link leads to the main content, the sidebar links and every deal card are real links and buttons that Tab reaches, and a card opens the deal drawer. Escape closes the deal drawer and every dialog. Dropping a card on a column is a mouse drag, but the deal drawer has a labeled Stage menu that changes the stage from the keyboard, and the dialogs do not move or restore focus.
Partly supported
Landmarks, named dialogs (Deal detail, New deal), labeled fields, and a health score with an accessible name are in place, and axe-core finds no serious or critical violation on 44 page views. Two smaller findings remain open. Nothing has been tested with a screen reader.
Partly supported
The first axe-core run found 25 serious color-contrast violations (health chips, accent buttons in dark mode, the delete text, dimmed settings rows). They are fixed, with source-level tests, and the rerun found none. Automated checks do not evaluate hover or focus states, so those are not covered.
Not yet tested
The mobile menu slides in over 200 milliseconds. Envoy's own styles have no prefers-reduced-motion rule, and nobody has checked whether the design system supplies one.
Not yet tested
Resizing text to 200 percent has not been tested.
Partly supported
Every page and the deal drawer fit a 375 pixel wide screen with no sideways page scroll, in both themes, and the sidebar becomes a menu button that opens a drawer. The kanban board scrolls sideways inside its own area. Tap-target size was not measured.
| Backend |
|
|---|---|
| Frontend |
|
| Scoring |
|
| For agents |
|
| Integrations in the code |
|
Envoy's MCP server exposes the whole product as tools an AI assistant can call. Each one is a thin call to the same REST API the web app uses, so nothing an agent can do is hidden from the app. The server runs over stdio and is not registered with the fleet's MCP gateway yet.
list_pipelinesget_pipelinecreate_pipelineupdate_pipelinearchive_pipelinedelete_pipelineseed_pipelineadd_stageupdate_stagedelete_stagereorder_stageslist_dealsget_dealcreate_dealupdate_dealdelete_dealrestore_dealchange_deal_stageget_deal_historycompute_deal_healthget_deal_risk_flagsrecompute_deal_risk_flagsget_deal_nbalist_accountsget_accountcreate_accountupdate_accountarchive_accountdelete_accounttribe_autocompletelist_activitieslog_activitydelete_activitysync_fleet_activitieslist_taskscreate_taskupdate_taskdelete_taskcompute_forecastsubmit_forecast_commitforecast_historycreate_quotelist_quotes_for_dealget_quotenew_quote_versionsend_quoteget_public_quoteaccept_public_quotedecline_public_quotecreate_sequencelist_sequencesenroll_in_sequencepause_enrollmentresume_enrollmentstop_enrollmentcreate_territorylist_territoriesassign_rep_to_territoryunassign_rep_from_territoryroute_leadcreate_commission_planset_quotaaccrue_commissionclawback_accrualget_commission_statementpost_commission_payoutrun_reportcreate_reportlist_reportsexecute_reportseed_reportscheck_service_healthget_diagnosticsget_my_identityNo. Envoy is in development and is not open to customers. To ask about it, write to contact@renkara.com.
No. Every screenshot is a capture of the running app against a made-up company, StrataForge Robotics. The deals, accounts, people, and numbers are fictional, and nothing was sent anywhere to take them.
No. The health score, the risk flags, and the next best actions are fixed rules over a deal's stage, activity, close date, amount, and contacts. The app labels the panel AI; the rules are on the scoring page, and the label is not accurate yet.
A placeholder. It is the weighted forecast multiplied by 1.1, and the Forecast page says it is a 10 percent uplift in mock mode. No model is behind it yet.
Not yet. Clients for Courier and Cadence are written, and they run against built-in sample responses by default. Switching them to the real services has not been done.
That is the aim, but it is not released, and the pieces above are what exist. Quotes are not on a screen yet and the fleet integrations are not live.
Parts of it are, and we say exactly which on the accessibility page. A deal moves by dragging on the board or from the Stage menu in the deal drawer, which works from the keyboard. The board's drop itself is still mouse-only, and that is listed as a known gap.
Write to Renkara Media Group and a person will answer.