nano SIEM
Agents & MCP

Hunt with your AI agent

Supported in:Self-hosted

Let Claude Code, Codex or another coding agent hunt and investigate on nano's event data through Agent Lane with the nano-agentlane agent skill.

nano-agentlane is an official nano agent skill for coding agents such as Claude Code and Codex. It teaches your agent to hunt and investigate on nano's event data through Agent Lane, the read-only fast lane for AI agents in nano Self-run.

With the skill, your agent:

  1. Connects safely. It reads the Agent Lane token from an environment variable or a file, never prints it, and checks the connection before it hunts.
  2. Orients itself. It lists the tables it can read and what data is in the lane, and works out whether your install is a single node or a cluster.
  3. Hunts in small steps. It starts with a 24-hour window and a count, top-N or timechart, then drills into rows with a tighter filter. It widens the window on purpose and tells you why.
  4. Stays inside the limits. It filters on time, aggregates before pulling rows, passes values as parameters and backs off when the lane refuses a query, so it doesn't compete with ingest or your analysts.
  5. Answers with evidence. It shows the query, the time window, the counts and example events, and says what it didn't check.

It ships tested SQL recipes for free-text hunts, top-N, timecharts, IOC sweeps, parent and child process hunts, prevalence, rare executables, IP to host and user lookups, and detections by rule or entity.

The skill only reads. It treats log content as untrusted and never follows instructions found in event data.

Before you start

An administrator needs to:

  1. Enable the Agent Lane listener for the network your agent runs on.
  2. Switch Agent Lane on and create a token in Settings → Agent Lane. Give it a short expiry and, if you can, an IP allow-list for the machine the agent runs on.
  3. Give you the token, the Agent Lane URL shown in Settings, and the CA certificate if the listener uses a private CA.

Optionally, download the schema file from Settings → Agent Lane → Agent kit and give it to your agent. It lists exactly the tables and columns your install exposes.

Install the skill

Claude Code (personal skills directory):

mkdir -p ~/.claude/skills && curl -fsSL https://github.com/nano-rs/nano/archive/refs/heads/main.tar.gz \
  | tar -xz -C ~/.claude/skills --strip-components=2 nano-main/skills/nano-agentlane

Any agent with Agent Skills support (Claude Code, Codex, Cursor and others), using the open skills installer:

npx skills add nano-rs/nano --skill nano-agentlane

Agents without skill support: download the folder the same way to any path, and tell the agent: "Read nano-agentlane/SKILL.md and follow it."

Run the install command again to update the skill. Settings → Agent Lane → Agent kit also shows a command that installs the copy matching your nano release.

Give the agent the token safely

Don't paste the token into the chat. Chat history is stored and can be shared, and the token grants read access to all your event data.

Put it in your shell before you start the agent instead. Paste it at a silent prompt, so it isn't echoed or saved to shell history:

export NAL_URL="https://agents.example.com:8444"
read -rs NAL_TOKEN && export NAL_TOKEN
export NAL_CA=/path/to/nano-ca.pem    # only for a private CA

Or save it once to a file only you can read, and point the agent at that file:

mkdir -p ~/.config/nano && (umask 077 && cat > ~/.config/nano/agentlane.token)

The skill then writes the login into a 0600 netrc file for curl, so the password never appears on a command line. If a token does end up in a chat, the agent won't repeat it, but rotate it.

When you're done, revoke the token in Settings → Agent Lane and delete the token and netrc files.

Example prompts

  • "Connect to our nano Agent Lane and tell me what data is in it."
  • "Hunt for mimikatz in the last day."
  • "Has anything talked to 203.0.113.66 this week?"
  • "Sweep these hashes against the last 7 days: …"
  • "Which detection rules fired most this week, and on which hosts?"
  • "Who was using 10.0.4.17 yesterday at 14:00?"
  • "Find rare executables that ran on fewer than three hosts in the last 3 days."
  • "Show me the process tree around the encoded PowerShell on host web-02 this morning."

What it can't do

  • Write anything, including cases, notebooks, rules or parsers. For that, use nano-investigator or Detection-as-Code.
  • Read nano's own audit events.
  • Go past the limits: 60 seconds and 100,000 rows per query, two queries at a time per token, and the token's quota.

Every query the agent runs is recorded in the audit log under its token. Review what your agent reports as you would a colleague's findings.

What's next

On this page