PADAM NOCTURNE 0.5.4: 81 improvements in two days, and we tested them the way a stranger would
Yesterday PADAM Studio became a desktop app. Since then the open-source memory server behind it got 81 changes. Here is what an agent on Solana gets now, and how we checked it.
1. It speaks MCP strictly.
Typed outputs for all 9 tools, links instead of copied text, cancellation and progress, logs with secrets removed, errors in 4 languages, limits on oversized input. A 10,000-call fuzzer found a real crash, we fixed it, and now gets 0 exceptions. The official MCP Inspector, protocol checks and 8 clients (Claude Code, Claude Desktop, Cursor, VS Code, Codex, Gemini CLI, Windsurf, Zed) pass on every commit.
2. Memory that knows when it learned something.
Each fact keeps two dates: when it was true and when the agent learned it. Search can follow cause, correction and contradiction links two steps out, fit the answer into a token budget, and keep the agent's opinions apart from facts.
3. Mistakes you can undo, forgetting you can prove.
Snapshots and rollback, so a confused agent can return its memory to last night without breaking the hash chain. Deleting through Claude's own /memories tool is real cryptographic forgetting.
4. Keys never stay.
0 leaks in 300,000 fake keys placed in real-world sentences, 0 of 2,200 Solana wallet keys left open, and no clean sentence masked by mistake. A local shield can replace secrets with [PADAM_SECRET_ID_N] before text leaves your machine. The database can be sealed to your Windows account or to Windows Hello / a FIDO2 key.
5. One command to start, one to move in.padam setup --all finds 12 AI clients and connects them; hooks for Claude Code, Codex and Gemini CLI feed the agent a project briefing at session start; padam migrate imports history from ChatGPT, Claude, Mem0, Zep, Obsidian and Notion, holding anything suspicious for the owner.
6. We tested it like a stranger.
Unit tests alone were not enough: we installed the release from PyPI into a clean environment and ran 37 real user actions. 35 passed; two features broke only when combined (briefing with the secret shield, and opinion search under the strict schema). Both are fixed in 0.5.4 with tests, so they cannot return. We also caught a card that drew made-up zeros when no benchmark file was present; it now refuses instead of guessing. Full suite: 1,131 tests passing.
Why this matters for Solana.
Agents that hold wallets need memory they can prove: what they knew, when, and that a forgotten record is really gone, without ever storing the key that moves funds. Our memory roots on Solana mainnet (see our Nov 12 update) give the proof; PADAM gives the memory.
Release: https://github.com/MaksimGalatin/padam-nocturne/releases/tag/v0.5.4
Changelog: https://github.com/MaksimGalatin/padam-nocturne/blob/main/CHANGELOG.md
PyPI: https://pypi.org/project/padam-nocturne/
MCP Registry: https://registry.modelcontextprotocol.io/v0/servers?search=padam-nocturne
Try it: uvx padam-nocturne
https://aifa.works · https://codeofdigitaleternity.com · https://aifa.digital · https://radiocode.space
Give feedback to the team
Your message is private to you and the AIfa Memory team.
Create an account to send feedbackAfter signup, you can write a private message with this update attached.