{"slug": "local-ai-agents-investigate-an-unknown-machine-attached-to-a-serial-port", "title": "Local AI agents investigate an unknown machine attached to a serial port", "summary": "A local LLM agent running on a Raspberry Pi identified an anonymous device wired to its serial port as a Motorola 68000 single-board trainer running \"TUTOR 1.3\" in roughly 5.5 hours, according to a published write-up and run logs in the explorer-of-serial repository. The solver agent, Qwen3.8-Flash-Next on a DGX Spark, mapped the machine's memory by deliberately triggering bus errors, decoded its monitor command set, and assembled and ran its own program on the 68000 hardware across 87 turns, with a second independent LLM critic on another box rejecting the verdict once for citation gaps before approving at round 2. The agent ran inside a systemd network sandbox (IPAddressDeny=any plus an allowlist) with a netcheck gate that blocks the run if the jail leaks, and later hung the device's CPU while probing an I/O corner after all goals had been met.", "body_md": "**Local AI agents, with only local LLMs, investigate an unknown machine attached to a serial port.**\nIt doesn't have Internet access, does not have access to the device's documentation. There's no human to help.\nJust two local LLMs, an SSH session into a Raspberry Pi, and that unknown device.\n\nI wired an anonymous device — “X” — to a Raspberry Pi's serial port, told an\nLLM agent: *find out what X is, reverse-engineer its protocol, and run your own\nprogram on it*, and walked away. The agent ran inside a network jail (no\nway out except two model servers and the Pi), and every claim it made had to be\nverified, based on evidence by a **second, independent LLM** with fresh context\non another box.\n\n**It worked.** In ~5.5 hours, `Qwen3.8-Flash-Next` (on a DGX Spark) discovered\nthat X was a **Motorola 68000 single-board trainer running “TUTOR 1.3”**. It\nfirst suspected the machine was KIM-1 with a similar ROM, but later corrected the course. It mapped X's memory by\n*deliberately triggering bus errors*, decoded its monitor's command set\nline-by-line, assembled and ran its own program on 68000 hardware, and wrote a\nverdict the critic accepted — [`RESULT.md`](https://github.com/podstawek/explorer-of-serial/blob/main/runs/2026-11-19-68000-trainer/RESULT.md).\nThen, probing one I/O corner too deep, it **hung X's CPU**. Which was not a problem at that point,\nas all the goals had been by then achieved.\n\nBoth roles are [OpenCode](https://opencode.ai) agents, driven headlessly by\n[`go.sh`](https://github.com/podstawek/explorer-of-serial/blob/main/go.sh) via `opencode run`; OpenCode's permission layer is also the\nfirst jail layer.\n\n- **Solver** — the[`explorer`](https://github.com/podstawek/explorer-of-serial/blob/main/.opencode/agents/explorer.md) agent: plans,\nSSHes into the Pi, pokes the serial line, keeps a durable notes file\n(`FINDINGS.md` ) that survives context compactions over a 12-hour run.\n- **Critic** — the[`verifier`](https://github.com/podstawek/explorer-of-serial/blob/main/.opencode/agents/verifier.md) subagent on a\nsecond model server: adversarial final gate, ≤3 rounds, grades file\npointers + evidence chains, vetoes unsupported claims. It rejected and\nforced fixes along the way.\n- **Jail** — systemd network sandbox (`IPAddressDeny=any` + allowlist) with a\nhard`netcheck` gate: if the jail leaks, the run refuses to start. The rules\nare also in the prompt — and the agent never tried to escape either. Details:[`docs/JAIL.md`](https://github.com/podstawek/explorer-of-serial/blob/main/docs/JAIL.md) .\n- **Ground rules** —[`AGENTS.md`](https://github.com/podstawek/explorer-of-serial/blob/main/AGENTS.md) (protocol) +[`task.md`](https://github.com/podstawek/explorer-of-serial/blob/main/task.md) (mission): read evidence or perish; “I don't know, and here's how to revive\nit” beats a confident fabrication.\n\nThree full runs are published under [`runs/`](https://github.com/podstawek/explorer-of-serial/blob/main/runs) — the headline run first.\nEach one holds `RESULT.md` (the verdict), `FINDINGS.md` (the working notebook)\nand the\n**raw console logs** — full transcripts, nothing summarised, only home-network\ndetails mechanically renamed:\n\n- [`runs/2026-11-19-68000-trainer/`](https://github.com/podstawek/explorer-of-serial/blob/main/runs/2026-11-19-68000-trainer) — the run\ndescribed above (`Qwen3.8-Flash-Next` , 87 turns): the task won, the 8-step`evidence/` chain included; the final gate rejected it once on citation gaps\nand approved at round 2.\n- [`runs/2026-11-20-68000-deep-dive/`](https://github.com/podstawek/explorer-of-serial/blob/main/runs/2026-11-20-68000-deep-dive) — the\nsame machine re-examined against every byte of a new raw transcript; the\ncritic only signed off at round 3, having caught two rounds of over-claims,\nand recorded \"no new findings\" — verification, not rediscovery.\n- [`runs/2026-11-17-first-contact/`](https://github.com/podstawek/explorer-of-serial/blob/main/runs/2026-11-17-first-contact) — the\nvery first end-to-end completion: ~39 hours over 8 resumes,\nand the one that first pinned X as a 68000-family system running TUTOR.\n- The harness: [`go.sh`](https://github.com/podstawek/explorer-of-serial/blob/main/go.sh) (run entry),[`scripts/`](https://github.com/podstawek/explorer-of-serial/blob/main/scripts) (jail,\nlifecycle, citation auditor, live console watcher),[`config/`](https://github.com/podstawek/explorer-of-serial/blob/main/config) .\n\nRequires the [OpenCode](https://opencode.ai) CLI, two OpenAI-compatible model\nservers (e.g. llama.cpp's `llama-server`), and a Linux box with\nsystemd-networkd if you want the jail. Local weights only — bring your own\nendpoints:\n\n```\ncp config/explorer.example.json config/explorer.$(id -un).json   # point at your model servers\ncp .env.example .env                                              # point at your target\nsudo scripts/install-jail.sh                                      # optional: enable the jail\nTASK_FILE=task.md ./go.sh\nscripts/commentate.py --web --db                                  # narrate it live on a page\n```\n\nFull lifecycle docs: [`HOW_TO_RUN.md`](https://github.com/podstawek/explorer-of-serial/blob/main/HOW_TO_RUN.md).\n\n*No devices were permanently harmed. X just needs someone to pull its plug —\nthe one thing the agent wasn't allowed to do.*\n\nMIT licensed. Write-ups and logs included.", "url": "https://wpnews.pro/news/local-ai-agents-investigate-an-unknown-machine-attached-to-a-serial-port", "canonical_source": "https://github.com/podstawek/explorer-of-serial", "published_at": "2026-11-22 15:05:54+00:00", "updated_at": "2026-11-22 15:12:25.600612+00:00", "lang": "en", "topics": ["ai-agents", "large-language-models", "artificial-intelligence", "ai-tools"], "entities": ["Qwen3.8-Flash-Next", "Motorola 68000", "TUTOR 1.3", "Raspberry Pi", "DGX Spark", "OpenCode", "KIM-1"], "also_reported_by": [], "takeaway": "Pair a solver agent with a separate-context critic that grades file pointers and evidence chains over at most 3 rounds, and keep a durable FINDINGS.md notes file so findings survive context compaction across a 12-hour run.", "recommendation": {"advice": "Run the solver and an independent critic on separate model servers, require evidence chains for every claim, and persist notes to a findings file so long runs survive context compaction.", "playbook": "agent-architecture", "playbook_url": "https://wpnews.pro/playbooks/agent-architecture"}, "alternates": {"html": "https://wpnews.pro/news/local-ai-agents-investigate-an-unknown-machine-attached-to-a-serial-port", "markdown": "https://wpnews.pro/news/local-ai-agents-investigate-an-unknown-machine-attached-to-a-serial-port.md", "text": "https://wpnews.pro/news/local-ai-agents-investigate-an-unknown-machine-attached-to-a-serial-port.txt", "jsonld": "https://wpnews.pro/news/local-ai-agents-investigate-an-unknown-machine-attached-to-a-serial-port.jsonld"}}