Wondering how to get Hermes Agent stream JSON output working? The quick version is just below — and honestly, this is one of those small flags that quietly changes how much you can automate with Hermes.
Short answer:
- Hermes Agent v0.21.4 (v2026.9.21), released 21 September 2026, adds a
--format stream-jsonflag to the CLI, per the official GitHub release notes. - It streams structured JSONL — one JSON object per line — instead of plain text, so scripts and pipelines can parse Hermes output reliably.
- You get it by updating:
hermes updatefor git installs, the installer one-liner, or thenousresearch/hermes-agent:v2026.9.21Docker image. - The same release rolls up roughly 1,812 merged PRs, including skills auto-load and configurable MCP discovery concurrency.
- It costs nothing extra — it’s a built-in feature of the open-source agent.
What Is Hermes Agent Stream JSON Output?
Hermes Agent stream JSON output is a new CLI output format that arrived in v0.21.4, tagged v2026.9.21 and released on 21 September 2026, per the official release notes on the NousResearch/hermes-agent GitHub repository. Instead of the normal human-readable session text, you pass --format stream-json and the CLI emits structured JSONL — a stream of JSON objects, one per line, as the session runs.
If you’ve only ever used Hermes interactively, that might sound like a nothing feature. It isn’t. Plain text output is for humans; structured output is for machines. The moment your agent’s output is JSONL, every line becomes something a script can parse without regex guesswork: pipe it into jq, log it, filter it, or hand it to another program that decides what happens next.
This is the same pattern serious agent CLIs have converged on, and it matters because Hermes has stopped being just a chat tool. It runs scheduled jobs, posts to channels, manages sessions and drives whole workflows. Structured output is the connective tissue that lets you build on top of all that.
How to Get the Flag: Update to v0.21.4
The flag only exists from v0.21.4 onwards, so step one is updating. The release notes give three supported paths:
- Git installs: run
hermes updateand you’re done. - Standard installs: re-run the official installer one-liner from the docs.
- Docker: pull the tagged image
nousresearch/hermes-agent:v2026.9.21.
I’ve covered the general update process before in my guide to updating Hermes Agent in two minutes — the process hasn’t changed, only the version number has. Worth knowing: v0.21.4 is a stability rollup tag that consolidates roughly 1,812 merged PRs and over 5,000 commits since v0.21.3, aimed at Docker images, Hermes Cloud and hosted deployments. The maintainers note the full curated release notes for this window will ship with v0.22.0.
🔥 Want this set up without the guesswork? Getting Hermes Agent’s stream-json output wired into a real automated SEO workflow is exactly the kind of build we walk through together. Inside the AI Profit Boardroom you get 3,700+ members, four live calls per week, daily tutorials, done-for-you templates and a 30-day roadmap, so you’re never stuck figuring this out alone. And if you’d rather talk it through first, book a free SEO strategy session and we’ll map out exactly where AI agents fit into your SEO workflow.
How to Use Hermes Agent Stream JSON Output in Your Automations
Here’s the practical bit. Once you’re on v0.21.4, add --format stream-json to your CLI invocation and treat the output as JSONL. Three patterns I’d actually use with an SEO workflow:
1. Log everything, grep nothing. Redirect the stream to a file and you’ve got a structured session log. When a scheduled content job misbehaves at 3am, you query the JSONL instead of scrolling raw terminal text trying to work out what the agent did.
2. Chain Hermes into a pipeline. JSONL is line-buffered, so a downstream script can react in real time — parse each line as it arrives, watch for the events you care about, and trigger the next step of the pipeline the moment the agent finishes a task. That’s how you go from “I run an agent” to “I run a system”.
3. Build dashboards and alerts. Because each line is a JSON object, feeding session activity into a monitoring stack becomes trivial. Pipe through jq to pull out the fields you want, then ship them wherever you track your automations.
One honest caveat: the v0.21.4 tag confirms the flag and its purpose (structured JSONL output), but the exhaustive event-by-event schema documentation is the sort of thing arriving with the curated v0.22.0 notes. My advice — the same advice for any streaming format — is to capture a real session to a file first, eyeball the objects it emits, and code your parser against what you actually see.
Stream JSON vs Plain Text Output
| Plain text (default) | --format stream-json | |
|---|---|---|
| Made for | Humans reading a terminal | Scripts, pipelines, wrappers |
| Parsing | Regex and hope | One JSON.parse per line |
| Logging | Unstructured text files | Queryable structured logs |
| Chaining tools | Fragile, breaks on format changes | Stable machine contract |
| Best for | Interactive sessions | Scheduled jobs and automations |
The rule of thumb: if a human is watching, keep the default. If a machine is watching, use stream-json.
What Else Shipped in v0.21.4
The stream-json flag is my pick of the release, but the 21 September tag bundles a lot, per the release notes:
- Skills auto-load: a
skills.auto_loadsetting pins your skills into every new session’s prompt automatically — pairs beautifully with stream-json for hands-off scheduled work. - Gateway hardening: a new
declineresponse setting for unauthorised DMs hitting the gateway. - MCP discovery concurrency: a configurable connect cap via
mcp.discovery_concurrency. - Session search upgrades: after/before time bounds plus an OR-relaxed recall retry.
- Desktop quality of life: chat font picker, one-click local engine updates, and plugin uninstall from the Plugins hub.
- New video catalogues: LTX 2.5 and Kling O3 added.
- A dozen new community plugins: Tailscale, SSH, Shodan, terminal, RSS and more.
If you’re still deciding whether Hermes deserves a slot in your stack at all, my full Hermes Agent review covers the fundamentals, and if you want it running against your Claude subscription without burning API credits, see Hermes Agent with a Claude Code subscription. For always-on channel automation, Hermes Agent bot mode is the natural next read.
The Bottom Line on Hermes Agent Stream JSON Output
Hermes Agent stream JSON output won’t demo well on a livestream, and that’s exactly why I like it. Flashy features get the views; structured output is what lets you turn an agent into infrastructure. Update to v0.21.4, flip the flag on your next scheduled job, and start treating Hermes sessions as data you can build on. Released 21 September 2026, free, and already in the tagged Docker image — there’s no reason to wait for v0.22.0.
Where to go from here: You now know what the flag does and how to switch it on — the leverage comes from what you build with it. Join the AI Profit Boardroom for the live calls, templates and the 30-day roadmap, or book a free SEO strategy session and I’ll show you where structured agent output fits in your funnel.
FAQ: hermes agent stream json output
What version of Hermes Agent added stream JSON output?
The --format stream-json CLI flag shipped in Hermes Agent v0.21.4 (tagged v2026.9.21), released on 21 September 2026. If you’re on v0.21.3 or earlier, you won’t have it — run hermes update first.
What does the stream-json format actually give you?
Structured JSONL output from the CLI: instead of human-readable text, the session streams machine-readable JSON you can parse line by line in scripts, pipelines and wrappers. That’s what makes Hermes usable as a component inside bigger automations rather than just an interactive chat.
How do I update Hermes Agent to get the flag?
For git installs, run hermes update. Otherwise re-run the official installer one-liner, or if you’re on Docker, pull the nousresearch/hermes-agent:v2026.9.21 image. The release notes recommend exactly these paths.
Is stream JSON output only useful for developers?
No — it’s useful for anyone wiring Hermes into an SEO or content workflow. If you want another script, dashboard or agent to react to what Hermes does, structured output is the difference between fragile text-scraping and a reliable pipeline.
Where are the full release notes for v0.21.4?
On the official GitHub releases page for NousResearch/hermes-agent, tag v2026.9.21. Note the maintainers say the full curated notes for this window ship with v0.22.0 — the v0.21.4 tag is a stability rollup of roughly 1,812 merged PRs.
Does Hermes Agent stream JSON output cost anything extra?
No. It’s a built-in CLI feature of the open-source Hermes Agent — you only pay for whatever model or engine you run Hermes on, exactly as before.
About the author
Julian Goldie is an SEO agency owner with 10+ years in SEO, 394K+ YouTube subscribers, a 100% Upwork job-success score, 75K+ community members across his groups, and the author of a best-selling SEO book. He tests AI tools and agents daily and publishes what actually works.
Watch the experiments on YouTube, learn AI SEO with 3,700+ members inside the AI Profit Boardroom, or book a free SEO strategy session to get a custom plan. For done-for-you work through the Goldie Agency, book a call for a custom quote.
Related reading
- How to Update Hermes Agent (2 Minutes)
- Hermes Agent Review
- Hermes Agent Bot Mode
- Hermes Agent + Claude Code Subscription: No API Key
Last updated September 2026. This is the living guide to hermes agent stream json output — it gets updated as the tools change.