Changelog #9

This covers everything shipped after Salestrics-AI-v2 was introduced as the platform default — from model selection through the intelligence layer, experiential polish, org customization, and launch assets.

Salestrics-AI-v2 foundation

Dual-model architecture

• Branded models — Salestrics-AI-v1 (fast, everyday) and Salestrics-AI-v2 (advanced reasoning); provider model names never shown in UI
• Platform default — Salestrics-AI-v2 set as default for new orgs
• Per-user picker — model choice in Assistant, persisted locally
• Admin default — Admin → AI Guardrails default model for embedded AI and workflows
• Migration — 0066_ai_default_model.sql adds default_model_id to org guardrails

Salestrics-AI-v2 runtime profile

• 8192-token completion budget (vs 4096 on v1)
• 10 tool-call iterations (vs 6 on v1) for multi-step investigations
• Higher embedded AI token defaults (3072 vs 1536) for app panels
• Model-specific temperatures and capability chips in the model selector
• v2 system prompt augmentations for multi-tool planning, cross-app synthesis, and structured action plans
• v2-aware contextual suggestions (admin vs staff prompt sets)

Public demo (ai.salestrics.com)

• Model selector added to signed-out demo chat
• Demo uses same v1/v2 branding and runtime profiles as the product
• Customer docs refreshed for trials, metering, and model selection

Compatibility & formatting

• Fixed v2 model compatibility issues in Assistant responses
• Improved paragraph formatting in streamed/markdown replies
• Fixed v2 empty responses on Custom Applications AI drafting — reasoning_effort: none for structured JSON, json_object response format, 8192-token draft budget
• Enhanced custom app drafting: data models, list views, relationships, automations, AI features, roles

Assistant UX upgrades

Messenger-style AI popup

• Global Assistant popup from utility bar / top bar — chat without leaving the current page
• Thread list + inline chat in a slide-over panel
• AssistantPopupProvider wired through app shell

Dashboard AI persistence

• Embedded AI panel exchanges on Dashboard and module landings now save as full Assistant sessions
• Links to reopen in popup or /assistant

UI cleanup

• Removed redundant top-bar AI trigger; popup is the primary quick-access surface
• Fixed popup layout and composer behavior

Intelligence layer (V2 → Intelligence)

Repositions Salestrics AI from chatbot to intelligence layer across the workspace.

Six branded experiences

Experience	Purpose
Assistant	Conversational AI with live org data
Briefings	Daily summaries and what-changed
Insights	Proactive observations
Actions	Workflow execution
Compose	Writing assistance
Analyze	Deeper business reasoning

Dashboard: Morning Brief

• Morning Brief card on org dashboard — “Good morning, {name}”
• Yesterday’s movement (leads, deals advanced, escalations, pipeline added)
• Today’s prioritized actions with estimated time
• Confidence badge (high / medium / low)
• One-click open in Assistant with pre-filled prompt

Dashboard: Salestrics AI Hub

• Central AI Hub on dashboard with experience tiles and prompt chips
• Inline flows: Explain my business, What changed?, Why is pipeline down?, Suggested automations
• Results render in-panel without leaving the dashboard

Dashboard: Insights Feed

• Proactive insights feed with dismiss
• Insight types: risks, wins, opportunities
• “I’m feeling lucky” — one surprising, actionable insight on demand
• Confidence levels and action prompts per insight

Assistant response intelligence

• Confidence indicators on replies (grounded in live data vs partial)
• Follow-up chips for natural next questions
• Action Mode buttons — one-click workflow triggers from responses
• Recommended next action extracted from assistant replies
• Record citations linking answers back to CRM records
• Rotating personality greetings replace static “Hello” (time-of-day + variety)

New query tools

Tool	What it does
org.what.changed	Compare today vs yesterday across deals, leads, revenue, activity
org.record.timeline	Chronological story across leads, opportunities, activities, cases

Workspace memory

• Org-scoped AI memory table (organization_ai_memory)
• “Remember this” facts persisted for future sessions
• Memory surfaced in intelligence flows

Connect meeting intelligence

• Automatic post-meeting AI follow-up after Connect video calls
• Summary saved to Assistant history

Database (migration 0070_ai_intelligence.sql)

• organization_ai_memory
• organization_ai_insights
• organization_ai_briefings

Experiential polish (“Wow in 60 seconds”)

First-run onboarding

• Role picker on first Assistant visit: Sales, Support, Founder, Operations, Finance
• Role tailors suggestions and idle prompts (stored per org in localStorage)

Investigation streaming

• Live “Investigating…” progress panel during tool execution
• Step labels: Reading CRM, Checking support, Reviewing finance, Building summary, etc.
• Steps stream over SSE as tools complete

Slash commands

Keyboard-first shortcuts in Assistant composer:

Command	Action
/brief	Morning brief
/pipeline	Deals closing + stalled
/summarize	Yesterday across CRM + support
/email	Draft follow-up email
/timeline	Account activity timeline
/analyze	Explain my business
/automate	Cross-workspace automation
/dashboard	Create Insight dashboard from intent

Cross-workspace automation orchestration

• Natural-language automation requests that span CRM, Mail, Connect, and Auto
• /automate maps intent → workflow draft → save

AI dashboard creation

• /dashboard creates live Insight dashboards from natural language
• Auto-detects focus: pipeline, support, operations, executive
• Returns dashboard name + deep link

Delight animations

• CSS animation system: ai-animate-in, ai-animate-slide-up, ai-animate-slide-right
• Applied to Morning Brief, AI Hub, Insights, and investigation UI

Idle prompt cycling

• Composer placeholder cycles through contextual prompts when idle
• Role-aware prompt rotation

Organization AI Instructions

Admin → AI Guardrails → AI Instructions (0071_ai_org_instructions.sql)

Configurable org context

Setting	Purpose
Organization context	Company, customers, operating model
AI role	Executive assistant, sales ops, finance analyst, etc.
Communication style	Concise, analytical, executive, friendly, technical, action-oriented
Business priorities	What AI should optimize for
Industry context	Compliance, procurement, domain nuance
Terminology overrides	Map your language (Customer → Account, etc.)
Restrictions	Topics to avoid or require approval

Admin experience

• Configuration status checklist
• Reset to Salestrics AI defaults
• Test AI configuration preview chat (Assistant, summaries, email drafts)
• Audit log (ai_org_instructions_audit) for all changes

Runtime injection

• Instructions injected into Assistant chat and all runAiCompletion surfaces
• Applies across Briefings, Insights, Analyze, Compose, Workspace, Mail, Orbit, embedded panels
• Never overrides platform safety rules or cross-tenant boundaries

Error hardening & reliability

• Centralized error codes (lib/ai/errors.ts) — 30+ ai_* codes with user-safe messages
• No OpenAI/OpenAPI/vendor details ever shown to users
• sanitizeAiErrorText() strips upstream leakage from raw errors
• Codes wired through: Assistant chat, demo chat, run-completion, guardrails, drafters, clients
• Docs: docs/troubleshooting.md — Salestrics AI error code section
• Internal: docs/internal/error-message-cheat-sheet.md — staff decode reference

Verification & tooling

• npm run verify:ai-intelligence — validates intelligence layer building blocks
• npm run verify:custom-app-ai — regression checks for custom app AI drafting
• RDS schema verify updated for new AI tables

Launch assets

Product Hunt launch trailer (/video)

• 45-second Remotion composition, 1920×1080 @ 60fps
• 9 cinematic scenes mirroring real Salestrics AI UI
• Piper TTS voiceover + FFmpeg ambient music + whisper.cpp captions
• One-command render: cd video && npm run render
• Output: video/output/salestrics-ai-launch-trailer.mp4

Infrastructure & build fixes

• Post-merge TypeScript fixes for availability narrowing, briefing types, demo client errors
• Excluded /video Remotion subproject from root tsconfig.json (separate package.json and dependencies)
• Deploy pipeline green on main with migrations 0070 + 0071

Summary: before vs after

Before (v1 only)	After (v2 + Intelligence Layer)
Single AI model	v1 + v2 with distinct runtime profiles
Chat-only Assistant	6 experiences: Assistant, Briefings, Insights, Actions, Compose, Analyze
Static dashboard	Morning Brief + AI Hub + Insights Feed
Generic replies	Confidence, citations, follow-ups, recommended actions
Manual navigation	Slash commands, investigation streaming, action buttons
One-size-fits-all prompts	Org AI Instructions + role onboarding
Raw error messages	Coded ai_* errors, zero vendor leakage
No cross-app memory	Workspace memory + proactive insights
No meeting intelligence	Connect post-meeting AI summaries
6 tool iterations	10 tool iterations, what-changed + timeline tools
No dashboards from chat	/dashboard creates live Insight dashboards
No automation from chat	/automate cross-workspace orchestration

Try it

Demo Salestrics-AI-v1 and v2 at ai.salestrics.com, read the Salestrics AI solutions page, or open the app for Morning Brief, slash commands, and org AI Instructions on your workspace.

Get support, email support@salestrics.com, or read docs.salestrics.com.