GitHub
I build most of these to challenge our own platform, pushing Personize's memory and governance into settings where the easy demo breaks down. Some are open source and production-tested. Others are experiments I'm exploring with a few engineers and domain experts, testing new architecture ideas before they graduate to public repos. More on github.com/personizeai.
Open Source
- revenue-osActive
The open-source OS for AI-powered revenue teams. 12 modules, 18 agent modes, multi-channel outreach across email, LinkedIn, and AI voice. Simulates a full sales org (SDR, AE, CSM, Ops, Analyst) with automatic handoffs between roles. Built to replace $5K/month SDR tooling with a transparent, governed pipeline you actually own.
- generative-sitesEarly Access
Drop a script tag, mark elements, get AI-personalized website copy per visitor. Property lookups, structured generation, and visitor deanonymization from a single <script>. The idea: every website becomes a 1:1 conversation, not a static page. Works behind any cache, CDN, or CMS.
- @personize/signalActive
A notification engine that decides IF, WHAT, WHEN, and HOW to notify. Scores every event against entity memory and governance rules, then sends, defers, or skips. Self-improving: captures open rates, click-throughs, and reactions to tune its own scoring over time. The goal is reducing notification volume by 40-70% while increasing engagement on what actually sends.
In the Lab
Private experiments and collaborative builds. Testing new architecture ideas with a small group of engineers and experts.
- content-osPrivate
Full content lifecycle OS: topic discovery, research, generation, quality review, multi-CMS publishing, repurposing, stale content refresh, and performance analytics. 12 modules, 6 editorial modes. One post auto-generates 5 derivative formats (LinkedIn, tweets, newsletter, email snippet, exec summary). Includes a prompt optimization eval framework that diagnoses and improves its own generation quality across all 10 pipelines.
TypeScriptContentAutomationAI - agents-uiPrivate
A control center for AI agent workflows. Agent templates, multi-provider engine selection (OpenAI, Claude, Gemini, Grok) with cost-per-record tracking, run management, and data source connectors. Early scaffolding for what a visual, no-code agent orchestration layer could look like. The bet: if you can see the agent graph, you can govern it.
TypeScriptUIAgent Orchestration - governed-memoryPrivate
250+ labeled samples across 15 experiments testing whether memory extraction, governance enforcement, and entity deduplication actually work under adversarial conditions. Conflicting policies, noisy inputs, ambiguous entity matches. The empirical receipts. If we are going to tell enterprises our governance is deterministic, we need the proof to back it up.
TypeScriptResearchValidationAI - ai-propertiesExploratory
Schema-driven AI property extraction for any entity type. You define the fields (text, number, boolean, date, options, arrays), and the system extracts and populates them from unstructured interactions automatically. AI generates the schema itself from a plain-language entity definition, then continuously fills and updates properties as new data arrives. The idea: your CRM schema should design itself and stay current without anyone manually tagging records.
TypeScriptMemorySchemaAI - segment-personizeExploratory
Turns raw Segment events (identify, track, page, group) into narrative memories that AI systems can consume safely. Allowlist-first policy filters low-signal telemetry before it hits memory. Events are translated into business-meaning narratives (support escalation, expansion signal, pricing interest) rather than stored as raw JSON. Three integration paths: Segment Destination Function (no server), webhook server (pilot control), and warehouse batch replay (historical backfill). Governance is first-class, not an afterthought, suppressing automation during support escalations and preventing tone-deaf outreach.
TypeScriptIntegrationsCDPMemory - smart-lead-scoringExploratory
Explainable, hybrid lead scoring that combines human-defined YAML rules with AI-extracted signals from Personize memory. 12 calculation types (semantic match, recency decay, range, regex, completeness, and more). Confidence-gated: AI extractions only count when they meet configured thresholds, so unreliable signals never inflate scores. Every score includes a full breakdown showing which criteria fired, points earned, data source, and confidence. Syncs grades and explanations back to HubSpot and Salesforce. Built to replace $800-5K/month scoring tools (Madkudu, 6sense, Einstein) with something transparent and self-hosted.
TypeScriptScoringSalesAI - ai-match-makerExploratory
Semantic matchmaking for conferences, referral networks, communities, and mentorship programs. Participants share interests via voice calls (ElevenLabs), web forms, or CSV. AI extracts what people offer, seek, and are exploring, then matches at the interest level, not the person level, so one person can match with five others on five different topics. Uses Personize similarRecords for embedding-based matching that captures meaning, not keywords. Cross-type scoring weights offering-to-seeking higher than peer-to-peer, surfacing the most valuable connections first. Matched pairs get personalized intro emails explaining the specific shared interest.
TypeScriptMatchingEventsAI - team-memoryExploratory
A suite of four Personize-powered skills (dev-memory, pm-memory, team-learning, team-memory-setup) that give AI coding agents persistent, shared knowledge across IDEs, sessions, and team roles. Five knowledge collections: files (pitfalls and conventions per source file), developers (expertise and preferences), features (spec and decisions), projects (architecture and conventions), and stakeholders. Works in Claude Code, Cursor, Windsurf, and terminal. The core loop: always check memory before editing, always save learnings after fixing, so the next developer (or agent) never repeats the same mistake.
TypeScriptDeveloper ToolsMemoryAI