Skip to content

Exporting to a SIEM and Signed Evidence

Omnodex can write what it recorded to a file your existing tools already know how to read, and it can prove that record hasn’t been edited since it was written. Both run from the CLI on the machine that did the recording, on every tier, and neither needs an account.

omnodex export writes events and findings from the local event log to a file, one JSON object per line. It opens no network connection. To get the events into a SIEM, point the collector you already run on the machine (a log forwarder or agent) at the file.

Terminal window
# Everything from the last week, in Omnodex's own form
omnodex export --out events.jsonl --since 7d
# OCSF, kept running: new events are appended, and the file rotates by size
omnodex export --format ocsf --follow --out /var/log/omnodex/events.ocsf.jsonl
OptionWhat it does
--out <file>Required. The file to write, or - for standard output. A one-off export replaces the file
--format jsonl or ocsfjsonl (the default) is Omnodex’s own event. ocsf maps each event to OCSF 1.3.0
--since <time>, --until <time>The window: a date or timestamp, or an age such as 30m, 24h or 7d
--followKeep running and append new events until you stop it with Ctrl+C
--include <groups>Add back content that’s left out by default: prompts, parameters, titles, errors, or all
--max-size <size>, --keep <n>With --follow: rotate the file at this size (default 50m) and keep this many old files (default 5)

Every Omnodex home the CLI reads on that machine goes into one export, oldest event first.

By default the export carries what happened, not what anyone typed or what a tool printed: prompts, tool parameters, session titles and error messages are removed. Add any of them back with --include. An event that had a field removed lists it under redacted, so a reader can tell a removed field from one the event never had.

Finding descriptions, file paths and working directories are always included. File contents never are, because Omnodex doesn’t record them.

With --follow, the export keeps a small <file>.watermark.json beside the output. If the command stops and starts again, it carries on from where it left off, so events aren’t skipped. If it stops at exactly the wrong moment, the last few events can appear twice, so have your SIEM deduplicate on the event ID.

Without --since, a follow starts with events recorded from now on. With --since, it first writes the history from that time. When the file reaches --max-size, it’s renamed to <file>.1 and a new one starts, which collectors that follow a file by name handle as an ordinary rotation.

Omnodex eventOCSF class
A tool call and its resultAPI Activity (6003), with the tool as the operation and the MCP server as the service
A file read or writtenFile System Activity (1001)
A findingDetection Finding (2004), with the rule as the analytic and the severity mapped across
A session starting or endingApplication Lifecycle (6002)
Anything elseBase Event (0)

Fields OCSF has no place for, such as the agent platform, the interceptor and the MCP server, are kept under unmapped. The class and activity numbers were checked against the published OCSF schema.

When you need to show someone that the record of what an agent did hasn’t been touched, export a signed bundle and let them check it with omnodex verify. They don’t have to trust you, or Omnodex.

Terminal window
omnodex seal key # this machine's key id and public key
omnodex export --bundle --out evidence.jsonl --since 30d
omnodex verify evidence.jsonl --key a7334e8a89ef5251

Omnodex hashes each recorded event into a chain, and signs each new stretch of the chain, called a seal, with a key that’s created on your machine and stays there. The background pass seals new events on its own, and omnodex seal does it on demand. Hooks do no extra work.

A bundle holds the seals for the range you asked for, the events under them, and the public key that signed them. Because a bundle is made of whole seals, it can cover a little more than the range you asked for. export --bundle seals anything new first, so a fresh bundle has nothing left unsealed.

omnodex verify checks every signature, recomputes every event’s hash and walks the chain:

VALID: 1 seal(s) and 849 event(s) check out.
Monitoring integrity events in range: none.

If anything was changed, it names what:

INVALID: 1 problem(s) found in 1 seal(s) and 849 event(s).
[edited] event sess_..._event_4 no longer matches the hash recorded for it

It reports events that were edited, dropped, added or reordered, seals that are missing or out of order, and signatures that don’t match. It exits 0 for a valid bundle, 2 when it finds a problem and 1 when it can’t run, so it fits in a script. --json prints the full report.

Pass the key you were given with --key, as its id or the full public key. Without it, verify can only say the bundle matches the keys inside it, which anyone who made the bundle could have chosen, and it prints those key ids so you can compare them yourself. Give whoever will check your bundles the output of omnodex seal key ahead of time, over a channel you trust.

verify also lists any capture pauses and monitoring changes inside the bundle, so a reader can see if recording was off for part of the range.

  • Events recorded after the last seal aren’t covered until the next seal.
  • Someone who controls the machine can delete the log and the key, or rewrite history and sign it again with the same key. A signature shows a record wasn’t changed after it was sealed, not that it was never changed by the machine’s owner.
  • A bundle holds every event in full, prompts and tool parameters included. Check it before you hand it over.

The checks run on Node’s built-in crypto with no other dependencies, and the format is documented in the open-source repository, so anyone can write their own verifier.