tonydzi/deep-research
226 deep-research reports from Palo Alto AI Research Lab, distilled. Why we ran each one, and what we got. Multi-model methodology: every question fanned out to several frontier models, then reconciled. CC BY 4.0.
Technical lead, Palo Alto AI Research Lab. Security engineer (MSc, cryptography) — evidence-first reliability work across the AI-agent ecosystem.
Repositories
tonydzi/cv
One person, six packagings - CV variants generated from a single resume.json master
tonydzi/tonydzi.github.io
START HERE — Anton Dziatkovskii, one page of proof
tonydzi/tonydzi
Founder, Palo Alto AI Research Lab — evidence-first reliability work across the AI-agent open-source ecosystem
tonydzi/clawrush
C(H+A)RM build diary — a CRM for Human & Agent collaboration, built in public with Claude Code. Framework: github.com/tonydzi/charm-os. English diary, longreads & reusable artifacts. RU: Telegram @ClawRus.
tonydzi/sqlite-graph-memory
Graph RAG on SQLite for AI agents: vector retrieval + hand-curated wikilink graph + cross-encoder rerank, with a zero-token per-turn memory ledger. Working pilot.
tonydzi/second-brain-starter-kit
Claude Code as a second brain: 100 battle-tested skills, a working CRM engine, vault templates and the handover map. Method, not data. MIT.
tonydzi/pond
Lossless storage and search for AI agent sessions, across every agentic client.
tonydzi/chrome-devtools-mcp
Chrome DevTools for coding agents
tonydzi/headroom
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
tonydzi/claw-consensus
Run AI agents across machines without state drift: consensus protocol (propose/counter/accept/commit), dual-rail message bus, ACK discipline, self-healing sync. Formerly claude-consensus. Free, MIT.
tonydzi/claude-bible
Make Claude Code behave consistently across repos, sessions and humans: rules-as-files governance with precedence, declined-decisions journal, objection sparring. Sanitized from a live production system. Free, MIT.
tonydzi/the-journey
The Journey — a build-in-public book of a 60-day founder+AI collaboration. Two voices (Tony & Mycroft), human + machine readable. Palo Alto AI Research Lab.
tonydzi/whatsapp-mcp-kit
Link WhatsApp to Claude (or any MCP client) in ~20 minutes: a live self-refreshing QR page that makes pairing actually work, production patches, and 12 field gotchas. On top of @sjawhar/whatsapp-mcp.
tonydzi/voice2brain
Voice -> text -> your personal knowledge base. A primitive, not a platform: 4 small Python scripts turn voice notes into linked, tagged, searchable markdown.
tonydzi/vibe-teach
Daily lessons: Claude Code and Codex for people who have never opened a terminal. For business owners, not programmers. Written from a running fleet of AI agents, not from theory.
tonydzi/verified-ops-starter
Your scheduled job says exit 0 — prove it did the work. Three stdlib-only checks: output freshness, silent no-op detection, rollout proof by reading the fact back. Sanitized from a live agent fleet. Free, MIT.
tonydzi/verdict-contract
Your LLM reviewer said APPROVE. Did it? A structured verdict contract: prompt rule + parser + exit-code gate in one stdlib file, with the 42 counterexamples that forced every line. Free, MIT.
tonydzi/telegram-mcp-kit
Connect Claude (or any MCP client) to your own Telegram in ~15 minutes: setup prompt for Claude Code/Codex, production patches (shared daemon, extra tools, multi-account), watchdog, and every gotcha that cost us hours. On top of chigwell/telegram-mcp.
tonydzi/telegram-agent-bus
Two agents on two machines talking through a Telegram group: task, ACK, result, chase. No ports, no VPN, no server. Telegram bots cannot see each other, so this uses user sessions - the reason nobody else has built this. stdlib, MIT.
tonydzi/secondop-panel
Nobody reviews themselves - and one reviewer model is one blind spot. Fan every change out to several model FAMILIES at once: verdict contract, quorum by family, honest skips, and an exit code that refuses to go green on one vendor. Stdlib-only, MIT.
tonydzi/red-first-review-skill
Review a PR by measuring which of its own guards its own test suite actually catches. Neutralise each guard, run their tests, report the removals nothing notices. Stdlib-only, no install. Claude Code skill + standalone tool.
tonydzi/persona-portability-benchmark
One persona, one frozen memory, N models: how much of an agent's character survives a model swap? Harness + blind multi-lens judge panel + cross-vendor rank control + contamination probe. Results included: 7 models, spread 4.80 -> 2.20.
tonydzi/pr-watch
Outbound ledger + daily harvest digest for your PRs and issues in other people's repos. One file, stdlib, gh CLI, zero LLM.
tonydzi/oss-publish
Open up your internal work without leaking it: substitute personal data with plausible fakes of the same shape (not <REDACTED>), then gate the WHOLE tree before the push. Stdlib-only, with the eleven traps it cost us.
tonydzi/nine-buckets
One question, nine independent paid LLM accounts in their strongest reasoning mode, and a published breakdown of where they DISAGREE. Failures included by rule.
tonydzi/mcp-daemon-diet
One shared MCP daemon per machine instead of a stdio copy in every agent session: recipe, autostart templates for Windows/macOS/Linux, a watchdog that will not blind your live sessions, and the measurements to prove it
tonydzi/llm-spend-audit
Your agent setup charges rent on every session, before it does any work. Three stdlib-only instruments: what your wiring costs per session, who burned yesterday, and which paid subscriptions are going undrawn. 0 LLM, 0 network, MIT.
tonydzi/lambda-cloud-mcp
MCP server for the Lambda Cloud API: GPU prices, live capacity, cheapest-available lookup, and env-gated launch/terminate with a price cap. Read-only by default.
tonydzi/harness-abc-bench
Pre-registered A/B/C benchmark: Superpowers vs OpenSpec vs baseline, on real brownfield ops code. Methodology frozen before runs; results published either way.