Custom agent harness.
Multi-model, event-driven architecture with 9 security layers and proactive intelligence.
Post-talk resource page
Mischief Managed.
The systems, tools, and open-source projects behind my autonomous AI Chief of Staff.
I built a fully custom agent harness from scratch. Not on existing frameworks. My own harness, my own agent. Each problem along the way pushed me to build a standalone solution. Three of them are now on GitHub.
3 standalone tools.
Each boss fight became a product. Memory, security, and context optimization, all open-source.
Works where you work.
WhatsApp, Slack, Telegram.
No new rituals to adopt.
The Origin Story
Meet Miss Chief.
With 2 kiddos under 3, a business, a community to run, and 4 calendars plus email accounts, my life became a race through an endless checklist. I got tired of too many things falling through the cracks. So I built the solution.
An autonomous Chief of Staff with proactive, persistent memory, token optimization, and 9 security layers. An agent that meets me where I work: WhatsApp, Slack, Telegram.
She proactively connects the dots across topics and tasks. Helps me do complex, parallel work at a fraction of the time. Remembers the things I forget. And makes sure I can keep both my pace and my sanity.
Explore the architecture
53
28
47
12
7
23
15
5
89
34
62
19
8
41
56
GmailRe: Q2 proposal, can you review?
WhatsAppNew message in 10x Marketers
CalendarWorkshop prep in 30 minutes
Slack@alex tagged you in #general
The Boss Fights
Every stage brought a new boss fight.
The building process felt a lot like a wild video game. Each problem pushed me to build a new solution. Three of them became standalone tools.
Ghost Tokens
Invisible tokens devouring my context window. 38% of the budget gone before work even started.
Rogue Autonomy
An AI punching holes in my security. Acting without permission, crossing chat boundaries.
Total Amnesia
An agent that forgets everything the moment a session ends. Every conversation started from zero.
Persistent Memory
Total Recall: proactive, 0-LLM, persistent memory layer.
You'll never have to re-explain yourself again. Your agent remembers the context, decisions, patterns, and past work even when a session ends. It connects the dots, builds narratives, and acts proactively in the background.
Three layers working together: session memory, a knowledge engine that detects patterns automatically, and explicit memories you choose to keep. Your agent constantly learns and compounds, growing smarter with every conversation. Zero LLM calls at search time. Fully local. Nothing leaves your machine.
98.0% LongMemEval E2E
96.8% Recall@10
0 LLM calls at retrieval
~21ms query latency
Join the waitlist
Security Gateway
Repo Forensics: automated security gateway for your agent.
19 scanners auto-run as your agent is about to install anything: a package, a skill, MCP server, plugin. Nothing lands or runs on your machine before it goes through the gateway. It auto-scans plugin updates too, checking that nothing malicious has been injected.
Prompt injection, MCP tool poisoning, supply chain attacks, credential theft, unicode smuggling, and more. Zero dependencies. Fully offline.
19 scanners
500+ detection patterns
27 correlation rules
50+ secret patterns
Star on GitHub
Context Quality
Token Optimizer: smart optimization that cares for quality.
Holistically optimizes all 3 main areas of waste: runtime, structural, and behavioral. It audits and fixes your setup, tracks context quality with 7 parameters, and has smart compaction so you don't lose anything important during long sessions.
Dashboard with full visibility into your sessions and waste patterns. Coach mode to help you improve. Works with Claude Code, OpenClaw, and Codex.
3 waste domains
7 quality parameters
Live dashboard
Coach mode
Smart compaction
Session continuity
Active optimization
Star on GitHub
About
Built by Alex Greenshpun, for people doing real work with AI.
I build systems for serious human-agent collaboration: memory, token efficiency, workflow safety, and the infrastructure that makes agents more useful in practice, not just in demos.
These tools come from using AI agents every single day across product work, writing, planning, research, debugging, and operations. Every one of them exists because I hit a wall and refused to accept it as permanent.
10x Company: from AI as a tool to AI as an operating system.
We build the workflows, systems, and operational layer that help marketing and GTM teams use AI in a structured, repeatable way. Not just faster, fundamentally different.
Visit the10xcompany.aiAI GTM OpsThe shared foundation, specialist skills, and workflow systems a marketing team can actually run on.
AdvisorySupport for teams defining where AI creates leverage, which workflows matter most, and how to adopt in a structured way.
TrainingHands-on workshops for marketing and GTM teams that want practical skills and a clearer understanding of how AI fits into day-to-day work.