SOC operations in plain language.
Search, triage alerts, work cases and notebooks, hunt, and build parsers. It reads first; its writes are things like cases, notebooks, dashboards and saved searches. Set up
Load nano's agent skill into Claude Code, Codex, Cursor or Antigravity. Tell it your daily volume and where nano should run: it sizes the hosts, checks them read-only, installs it and proves that search works. Then the same agent onboards your log sources and works in nano with you.
Your coding agent runs on your workstation. nano's tools give it a procedure to follow and a narrow way into nano: an agent skill to deploy, a script to onboard log sources, and two MCP servers to work in the product.
On your machine
Loads nano's skill and MCP servers. Uses your SSH keys, kubectl context and cloud CLI login.
nano-deploy skill
nano-connect
nano-investigator + nanodac
Logs are collected, stored and searched here and never sent to nano. Inside the product, pivt helps with queries, detections, parsers and case investigation. It is off until you point it at your own model endpoint.
nano-deploy is an Agent Skill: a folder with instructions, references and scripts your agent reads when you ask it to install, size, upgrade or troubleshoot nano.
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-deploy npx skills add nano-rs/nano --skill nano-deploy The open skills installer asks which agents to install into. For an agent without skill support, download the folder the same way and tell it: "Read nano-deploy/SKILL.md and follow it." Run the command again to update the skill. Deploy with your AI agent
The skill walks your agent through intake, sizing, a read-only preflight, the install and a verification, and stops at a gate after each phase.
Deploy nano on nano-1.internal as a single node. About 20 GB/day, 365 days retention. My enrollment values are in ~/nano/enrollment.env.
Deploy nano to these 8 VMs, about 12 GB/day, one-year retention. The VM list is in hosts.txt.
Plan nano on new VMs in AWS for about 1 TB/day. Show me the plan and the cost range before you create anything.
Is there a new version of nano? Upgrade it if nothing blocks the upgrade.
It asks for your hosts and SSH user (or your Kubernetes cluster), volume, retention, hostnames and TLS, and sizes the hosts from your GB/day. For several VMs, its planner lays out the roles so that losing any one VM keeps every quorum and a copy of all data, and refuses when the VMs are too small.
Over SSH it checks the OS, CPU and AVX2, memory, disk, Docker, time sync, ports and connectivity between hosts. On Kubernetes it checks with kubectl: cluster and Helm versions, storage class, Gateway API, node pools, and your external PostgreSQL. Any FAIL stops the deployment until you approve a fix.
On servers it downloads the release with your license values, checks its signature and checksum, then installs with Docker Compose in the documented order. On Kubernetes it pulls nano's Helm chart, with digest-pinned images, and runs helm install. Before each step it shows the commands and hosts and waits for your yes.
Health endpoints, admin login, license state, and a unique test event ingested and found by search. On HA it also checks the Keeper quorum and every ClickHouse replica.
| Have ready | Where it comes from |
|---|---|
| License values | app.nano.rs, the deployment you registered. Save them to a file (mode 0600) and give your agent the path. |
| Hosts | Linux hosts with a non-root SSH user that has sudo. The agent uses your SSH agent or key file. For new cloud VMs: your logged-in aws, az or gcloud CLI. |
| Hostnames and TLS | An app hostname, an ingest hostname, and a certificate or load balancer. |
| Volume and retention | GB/day or events per second, and how many days to keep. One server handles up to about 100 GB/day. |
Every install starts with zero log sources. nano-connect is a Bash script that turns the sources you pick into running collectors, and your agent can drive it end to end.
curl -fsSL https://github.com/nano-rs/connect/releases/latest/download/connect.sh -o connect.sh
bash connect.sh --version
export NANO_URL="https://nano.example.com"
export NANO_API_KEY="..." # from your secret manager
bash connect.sh list-sources
bash connect.sh add-source --sources cisco_asa,palo_alto:tcp \
--deploy-parsers --out-dir ./fw-collector --non-interactive
bash connect.sh connect --source cisco_asa --non-interactive What it does. Lists the sources your deployment's parser catalog offers, imports and publishes the parser for each source you pick (with --deploy-parsers), issues one ingest token per source, and writes a Vector collector for syslog devices, Linux journald and files, or Windows Event Log and Sysmon.
How it proves it. connect sends a marked test event and waits until search finds it under that source. verify checks that real events are arriving.
What it needs. Bash, curl and an API key with only the permissions it uses: log_sources:view to discover, settings:system to issue tokens, the parser permissions for --deploy-parsers, and search:execute to prove delivery. Tokens stay in private files and never appear on screen. nano-connect docs
Two MCP servers let your agent search, triage, investigate, hunt, build parsers and manage detections as code, through nano's API and with a key you scope.
Search, triage alerts, work cases and notebooks, hunt, and build parsers. It reads first; its writes are things like cases, notebooks, dashboards and saved searches. Set up
Draft detections as YAML in Git, validate and test them against real data, and sync them to nano through pull requests. A CLI, an MCP server and a GitHub Action. Set up
Drafts queries, detections, dashboards and parsers, tunes rules and investigates new cases. Included with every license, and off until you point it at your own OpenAI-compatible or Anthropic endpoint. pivt
{
"mcpServers": {
"nano-investigator": {
"command": "npx",
"args": ["-y", "@nano-rs/investigator-mcp-server"],
"env": {
"NANOSIEM_API_URL": "https://nano.example.com",
"NANOSIEM_API_KEY": "${NANOSIEM_API_KEY}"
}
},
"nanodac": {
"command": "npx",
"args": ["-y", "@nano-rs/dac-mcp-server"],
"env": {
"NANOSIEM_API_URL": "https://nano.example.com",
"NANOSIEM_API_KEY": "${NANOSIEM_API_KEY}"
}
}
}
} The agent's part is quick. What takes time is everything around it that belongs to your organization, and no agent can hurry that.
| Stage | Who | What sets the pace |
|---|---|---|
| Register a deployment | You, at app.nano.rs | Choose your daily volume. Trials are reviewed by our team first. The license values are shown once: save them to a file your agent can read, not into the chat. |
| Hosts | You. For full HA on AWS, Azure or GCP, your agent can create the VMs with your logged-in cloud CLI after you approve the plan | Whatever buying or approving a server takes in your organization, and cloud quota. New cloud projects often need a vCPU or disk quota increase. |
| DNS and TLS | You | Hostnames for the app and for ingest, and a certificate or a TLS-terminating load balancer. |
| Preflight, install, verify | Your agent, with your approval at each step | How quickly you approve each step, and fixing anything preflight marks FAIL (for example an old Docker version). The agent stops until it is fixed. |
| First log sources | Your agent with nano-connect; you change the devices | Access to the devices and endpoints, firewall rules and change windows. Pointing a firewall at a collector is a change on your network, not in nano. |
These are written into the skill your agent follows. They don't replace your review: read what it proposes before you say yes.
Before every step that changes something, the agent names the exact commands and hosts and waits. Destructive commands, such as deleting volumes or restoring a backup, need their own explicit yes.
The preflight checks, over SSH or kubectl, are read-only, and any FAIL stops the deployment. The agent never weakens security to get past a check: no installer run with sudo, no skipped image verification, no skipped backups.
License token, registry credential, admin password, .env and keys are moved as 0600 files and read on the host. The agent never asks you to paste one, and never asks for or stores cloud credentials.
nano's tools run on your machine and talk to your nano and your hosts. Your logs stay in your environment and are never sent to nano.
Your coding agent's model provider receives what the agent sends it: your prompts, the output of the commands it runs, and the results of the tools it calls. With nano-investigator, that includes search results from your logs. The skill keeps secrets out of that output.
If that is a problem, run the agent against a model inside your network, or one your policy approves, and scope the nano API key to what the task needs.
nano itself receives a license check-in and one number per day, the GB ingested. Never log data. What your install sends to nano
The nano-deploy skill uses the open Agent Skills format. Install it into Claude Code with the curl command, or into Codex, Cursor, Google Antigravity and other agents with the open skills installer (npx skills add nano-rs/nano --skill nano-deploy). An agent without skill support can read SKILL.md and follow it. nano-investigator and nanodac are MCP servers, so they work with any MCP client. nano-connect is a Bash script that any agent, or you, can run.
You tell it. For one server, a 3-server cluster or Kubernetes, you give it hosts or a cluster you already have. For a cluster with no single point of failure, its planner can lay nano out on your existing VMs or pick new VMs on AWS, Azure or GCP, with instance and disk types and a dated monthly cost estimate. Your agent creates cloud VMs only after you approve the plan, using your own logged-in aws, az or gcloud session.
For one server you already have, with SSH access, hostnames and a certificate ready, plan on an afternoon, not weeks. The agent's part (sizing, read-only preflight, install and verification) is the quick part. Buying or approving hosts, DNS, TLS, firewall changes and access to your log sources take as long as they take in your organization.
The skill tells the agent to show the exact commands and hosts before every step that changes anything, and to wait for your yes. One yes covers one step. Destructive commands need their own confirmation. Preflight is read-only and any failed check stops the deployment. Review what your agent proposes as you would a colleague's change: you stay responsible for your hosts.
No. Your logs stay in your environment, and nano never receives them. Your coding agent sends your prompts, the output of the commands it runs and the results of the tools it calls, including search results from nano-investigator, to the AI provider it is configured with. Choose that provider, or a local model, to suit your policy.
No. nano-connect is a Bash script that generates Vector collectors, issues one ingest token per source and checks that events arrive under that source in search. Every flag works non-interactively, which is what makes it easy for a coding agent to drive. It never runs an installer on its own: collectors start only when you run them or pass --run.
nano-deploy installs the paid product, so it needs the license values from the deployment you registered at app.nano.rs. The paid license includes every feature, pivt included; only the GB/day and support vary. nano-connect, nano-investigator and nanodac also work with nano's free open-core edition (AGPL-3.0), which doesn't include pivt.
Partly. Air-gapped installs are included with every license and use a signed offline license, which the skill knows how to import. The skill's download step reaches api.nano.rs, so on a network with no egress follow the Air-gapped deployment guide for getting the bundle and images across. Your coding agent also needs a model it can reach, such as one you run inside the network.
Last reviewed 13 November 2026.
From US$149/month at 5 GB/day, US$279/month at 10 GB/day (list price, billed monthly; 10% off billed annually). Every feature in one license, and one license covers two installs.