status

Mizukara is a local-first on-chain watchdog: it scans tokens against a set of verifiable rules and records every decision in a tamper-evident log on your own machine. This page is the one-stop summary of what ships today and what's on the roadmap.

Available today

• Token scanning — mizukara scan <address> runs five checks (R1–R5) against live data on chain 4663: source verification, ownership, mint authority, supply concentration, and liquidity. Every fact cites the exact query that produced it and a SHA-256 hash of the response, and a scan pinned to a block is reproducible byte-for-byte. See the rule reference for what each check does.
• Verifiable decision trails — every session writes an append-only, SHA-256 hash-chained log. mizukara replay <trail> verifies the chain: edit even one character of a trail after the fact and replay catches it instantly.
• Local gateway — a private daemon that runs on your machine only (loopback-only unless you explicitly expose it, bearer-token auth). It routes each chat message to a model tier and logs every routing decision to the trail.
• Private memory — a SQLite full-text store with keyword recall. Nothing in it ever leaves your machine.
• Sandboxed tools — file, shell, web, and git tools behind an approval-policy gate, with a working-directory jail for file/shell and a read-only allowlist for git.
• Simple onboarding — mizukara onboard sets everything up. API keys are never written to disk; run --ollama-only to stay fully offline.

Coming next

• Full chat — model replies from Anthropic, OpenAI, OpenRouter, and Ollama through the gateway, on top of the routing and logging that already run today.
• Deeper rules (R6–R9) — honeypot simulation (can this token be sold, not just bought?), deployer history, post-launch permission drift, and proxy hygiene.
• Automated monitoring — new-deploy detection and scheduled re-scans of known tokens over time, instead of one-off manual scans.
• Semantic recall — on-device embedding search for memory, alongside the existing keyword recall.
• Public findings — a findings index, a trail archive anyone can replay, and automated posting of new findings.
• Wallet-gated access — sign-in with your wallet for on-demand scans.

Read next

• Guide — get scanning in two minutes.
• Rules R1–R5 — exactly what each check verifies, with real examples.
• CLI reference — every command and flag.
• Architecture — how the trail, artifact, and rule engine work.