Hermes Fable 5: Set Up Claude Fable On 2 Routes (2026)

Hermes Fable 5 means running Hermes Agent with Anthropic's Claude Fable as the brain, and there are exactly two documented ways to wire it up.

The first is the official experimental plugin claude-subscription-directsdk, which runs on your Claude plan through Claude Code.

The second is Hermes's own built-in Anthropic provider, using an API key or a Claude login.

Below is how I'd set up each one, which plan rules apply, and how to stop Fable from running jobs that a cheaper model can handle.

Want my team to look over your agent setup with you? Book a free SEO strategy session and bring your config.

The Two Routes Side By Side

Pick your route before you type a single command.

	Plugin route	Built-in Anthropic route
What it is	Official experimental plugin that drives the Claude Code CLI	Hermes's native Anthropic provider
Login	Claude Code login (`claude auth login`), no API key	`ANTHROPIC_API_KEY`, or Anthropic OAuth through `hermes model`
Fable 5.1 id	`fable` → `claude-fable-5-1[1m]`	`claude-fable-5-1`
What pays for Fable	Your Claude plan (included on Max and Ultra, usage credits on Pro and standard seats)	API key: billed per token to the key's org. OAuth: only purchased extra-usage credits on Max
Status	Experimental review build	Core Hermes provider

The plugin is the one that can lean on the Fable allowance already inside a Max or Ultra plan.

The built-in route is the one to reach for when you'd rather keep agent traffic on an API key.

Check These Four Requirements First

The claude-subscription-directsdk README lists what the plugin route needs.

Hermes Agent 0.21.4 or newer.
Python 3.10 or newer on Linux, macOS or Windows.
The official Claude Code CLI, installed.
That CLI logged in to your Claude account.

On an older Hermes the plugin refuses to load and logs a clear Failed to load user provider plugin warning.

That's better than a half-working provider, but it means you should run hermes update before anything else.

Route 1, Step 1: Install Claude Code And Log In

The plugin has no credentials of its own.

Everything goes through the claude binary, so that comes first.

npm install -g @anthropic-ai/claude-code
claude auth login
claude update

The README says an outdated CLI answers with a 400 error that the model is not supported.

That's why I add claude update to the install every time.

If claude lives somewhere unusual, point CLAUDE_SUBSCRIPTION_DIRECTSDK_COMMAND at the binary.

Route 1, Step 2: Install The Plugin And Pick It

Install from the Hermes plugin catalog, which the README calls pinned and reviewed.

hermes plugins install claude-subscription-directsdk
hermes model

In the picker, choose Claude Subscription DirectSDK (Experimental).

The setup asks the CLI three things before saving anything: is claude installed, is it logged in, and which models does this account get.

If you want the version that tracks main instead, install NousResearch/hermes-official-plugins#claude-subscription-directsdk from GitHub.

The plugin lives in the hermes-official-plugins repo, next to the other official Nous plugins.

🔥 Want Fable wired into a full agent stack, not just a chat window? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members running Hermes setups like this one. → Join the AI Profit Boardroom

Route 1, Step 3: Make Fable The Default In config.yaml

You can pick Fable in the menu, or write it straight into ~/.hermes/config.yaml.

model:
  provider: claude-subscription-directsdk-experimental
  default: fable

Short names sonnet, haiku, opus and fable resolve to pinned routes.

fable lands on claude-fable-5-1[1m], which the README pins with a 1,000,000-token window.

For a one-off session, pass the provider on the command line instead.

hermes --provider claude-subscription-directsdk-experimental -m fable

Hermes keeps its own agent loop, tools, approvals and compaction on this route.

The README is clear that the plugin never opens, copies, refreshes or prints Claude Code's credential files.

Which Plans Include Fable On The Plugin Route

This is the part people get wrong.

Commit fdbf4225d2 on 2 November applied Anthropic's plan rule to both Fable 5 and Fable 5.1.

Max: Fable is included in the plan.
Ultra: Fable is included in the plan.
Pro: Fable runs on usage credits.
Team and Enterprise: the picker reads "usage credits on standard seats", because premium seats include Fable and the handshake can't see the seat tier.

That matches Anthropic's own help article, Claude Fable models on your plan.

An earlier commit, 4220bbaadb, made the picker list your account's own models from Claude Code's served catalog.

So the list you see is what your account can actually run, with a dim usage-credits note on any row that bills that way.

Watch /usage And Switch Off Extra Usage

Run /usage inside Hermes to see your plan limits.

The plugin asks Claude Code itself for those numbers, so it never reads or sends your login.

Then go into your Claude account and turn off extra usage if you don't want overage.

The README recommends exactly that, and it adds that a list-price estimate in Hermes is not proof of what your plan charged.

One setting can surprise you here.

claude_code_telemetry is on by default because it fetches the feature flags that unlock models in the picker.

Turn it off and the picker can shrink to each family's current model, and /usage tells you it can't fetch limits.

Route 2: The Built-In Anthropic Provider

Hermes ships a native Anthropic provider, documented under Anthropic (Native) in the providers guide.

With an API key, the setup is two lines.

export ANTHROPIC_API_KEY=your-key
hermes chat --provider anthropic --model claude-fable-5-1

The docs say this is billed per token to the key's organization, independent of any Claude subscription.

The OAuth option through hermes model reads very differently.

It only works on a Claude Max plan with purchased extra usage credits.

Hermes never touches the base Max allowance on that path, only the extra credits on top.

Claude Pro accounts can't use the OAuth path at all, so Pro users go with the API key.

The Hermes docs don't mention Ultra, Team or Enterprise for this route, so I wouldn't assume anything about them.

Fable 5.1 vs Fable 5: What Anthropic Says

Anthropic released Claude Fable 5.1, id claude-fable-5-1, on 1 September 2026.

Its model page lists a 1M-token context window, 128K max output and adaptive thinking that is always on, with default effort set to high.

Claude Fable 5 is now marked legacy, and Fable 5.1 is the current Fable model.

On choosing, the models overview says to start with Claude Opus 5.5 for most workloads.

It points to Fable 5.1 for demanding reasoning and long-horizon agentic work, or when Opus 5.5 at higher effort still falls short.

That's the brief for Hermes too: Fable is the planner, not the clerk.

Keep Fable On The Main Chat, Not Every Subagent

By default, subagents inherit the parent's model.

That means every delegate_task child would also run on Fable.

The delegation docs let you route children to a cheaper model while the planner stays put.

model:
  provider: claude-subscription-directsdk-experimental
  default: fable
delegation:
  provider: claude-subscription-directsdk-experimental
  model: haiku

The plugin README says auxiliary and fallback routing stays owned by Hermes, so set these routes explicitly.

The Hermes FAQ also warns that each /model switch resets the prompt cache.

Delegation avoids that, because each subagent starts with its own fresh context.

Pin Your Cron Jobs Off Fable

Cron resolves a job's model in this order: per-job pin, then cron.model, then your main model.

So the moment Fable becomes your main model, every unpinned job runs on it.

The cron docs give two fixes.

hermes config set cron.model_provider claude-subscription-directsdk-experimental
hermes config set cron.model haiku
hermes cron edit <job_id> --provider claude-subscription-directsdk-experimental --model sonnet

The first two lines set a fleet default for every unpinned job.

The last line pins one job to an explicit model.

Hermes's own code calls Opus and Fable unsafe silent defaults for exactly this reason, in the frontier-tier check in models.py.

Three Gotchas That Stop The Setup

Gotcha one: a leftover API key in your shell.

The plugin rejects inherited API-key, custom-endpoint and cloud-backend overrides before it starts, and names the variable that clashes.

Unset it in the launching environment and run the request again.

Gotcha two: "Fable 5.1 requires usage credits".

The README logged that exact error on its own test account, and calls it Anthropic's plan rule rather than a routing bug.

The plugin passes the error through and never falls back to another model, so either add usage credits or pick another model.

Gotcha three: Hermes under launchd or systemd says you're logged out.

The README explains that Claude Code checks the process identity, and the plugin now fills in USER for you.

If it still fails, the error carries the CLI's first stderr lines, so read those first.

📞 Stuck between the two routes? Book a free SEO strategy session and my team will walk through which route fits the way your business uses Hermes. → Grab a free strategy call

Optional: Put A Ceiling On A Hung Call

One setting is worth adding once the basics work.

The README notes that Hermes treats the plugin's process:// endpoint as local, so its stale-call watchdog can stay off for some model ids.

A native call that hangs without output can then hold the turn for the whole read-idle limit before Hermes retries or falls back.

To bound it, set a stale timeout for the provider.

providers:
  claude-subscription-directsdk-experimental:
    stale_timeout_seconds: 900

On this path the value limits the whole call, not just a silence, so pick a number above your longest healthy turn.

Long Fable planning turns are exactly the ones that run long, so don't set it tight.

Two Setups From Small Businesses

A plumbing firm's morning job board.

A cron job on haiku reads the overnight booking emails at 7am.
It sorts them into urgent call-outs and routine visits.
It posts the list to the team chat.
The owner opens the main Hermes chat on Fable only when a tricky multi-site job needs a plan written out.

A three-person marketing agency's client audit.

The account lead opens a Fable session and pastes in the client brief and last quarter's notes.
Fable writes the audit plan and splits it into page-level tasks.
Each task goes to a sonnet subagent through the delegation route.
Fable merges the subagent findings into one audit document for the lead to review.

In both cases Fable does the thinking, and the lighter models do the repeat work.

Which Route I'd Pick

If you're on Max or Ultra, use the plugin route, because Fable is already part of the plan.

If you're on Pro, the plugin works but Fable bills to usage credits, so weigh that against an API key on the built-in route.

If you run production agents for clients, remember the plugin calls itself a review build, not a production-readiness claim.

In that case the built-in Anthropic provider with an API key is the steadier choice.

For picking the rest of your model lineup, my best Hermes Agent model guide covers the other options.

Want the full Agent OS with Fable already slotted in? Join the AI Profit Boardroom and copy the setup.

Also On Our Network

FAQ: Hermes Fable 5

What is the quickest way to run Hermes on Claude Fable?

Install and log in to Claude Code, run hermes plugins install claude-subscription-directsdk, then pick Claude Subscription DirectSDK (Experimental) in hermes model and choose fable.

Which model id does the plugin use for Fable 5.1?

The short name fable resolves to the pinned route claude-fable-5-1[1m], which the plugin README lists with a 1,000,000-token context window.

Why does Hermes say Fable 5.1 requires usage credits?

That is Anthropic's plan rule surfacing through Claude Code. On Pro and on standard Team or Enterprise seats, Fable runs on usage credits, and the plugin passes the error through without switching models.

Can I run Fable through Hermes with an API key instead?

Yes. Set ANTHROPIC_API_KEY and start Hermes with --provider anthropic and --model claude-fable-5-1. The Hermes docs say that path is billed per token to the key's organization, separate from any Claude plan.

Do my cron jobs switch to Fable when I change the main model?

Unpinned ones do, because cron falls back to the main model. Set cron.model or pin each job with hermes cron edit --model to keep them on a lighter model.

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
I've grown a YouTube channel to 400,000+ subscribers.
I wrote "Link Building Mastery".
My Udemy courses have taught over 29,000 students.

→ Get my best AI training inside the AI Profit Boardroom

Install the plugin, set default: fable, pin your cron jobs, and if you want a second opinion book a free strategy session before you run your Hermes Fable 5 setup on real client work.