Hermes Agent with BazaarLink: Custom Endpoint Setup and Free-Tier Limits

To connect Hermes Agent to BazaarLink, configure a custom endpoint with a base URL, your API key and a model ID. Installing Hermes and choosing the endpoint are separate steps. Setting a few generic OpenAI environment variables is not a substitute for Hermes's current provider configuration.

This guide was checked against the Hermes installation documentation and provider reference on 17 September 2026. The configuration below is documentation-based; this revision did not execute a live Hermes task through BazaarLink.

Start with one chat request, not an unattended learning loop

Hermes is Nous Research's agent software. It can run tasks and reuse skills, but a usable agent backend needs more than an HTTP endpoint: model capabilities, context capacity and enough call budget matter.

BazaarLink offers OpenAI-compatible chat access at https://api.bazaarlink.ai/v1. Its auto:free pool has a limited shared quota. Use it for a small connection check; do not treat it as an unlimited backend for hundreds of skill-refinement calls. A paid model or local backend is a better starting point if the agent needs sustained tool use.

Install Hermes using its supported installer

Choose the installer for your operating system from the official installation page. The documented CLI paths include Linux/macOS/WSL2 and native Windows; desktop installers are also available. Review a downloaded script before running it, particularly on a shared workstation.

After installation, open a new terminal and check that the CLI is available:

hermes --help
hermes doctor

The older recipe in this article—cloning a repository, installing requirements.txt and launching python run.py—was not a verified current installation path. Use the supported installer or the linked developer-installation instructions instead.

Create a BazaarLink key and choose the model

Sign in at /login, open /keys and create an API key for this installation. Store it privately. A separate key per application makes revocation and usage review easier; do not paste it into a public issue or a configuration example.

Browse /models for the exact model ID, supported features and context limit. auto:free is suitable for testing basic text access, but its selected model can change. For tool-using Hermes tasks, choose a specific model and verify tool-call support rather than relying on a model-family name.

Configure the custom provider through the wizard

Run this in your terminal, outside an active Hermes chat:

hermes model

Choose Custom endpoint and supply:

The wizard saves model/provider settings in ~/.hermes/config.yaml. Secrets have separate storage under the Hermes home directory. The current configuration reference explains precedence and storage.

Do not declare a larger context window just to get past startup validation. Hermes's provider documentation warns that tool-using agents require sufficient context; a setting cannot enlarge the upstream model's real capacity. If automatic discovery fails or the selected free model is unsuitable, verify a specific model in the catalog before continuing.

Inside a running chat, /model switches already configured providers/models. It does not perform the full new-provider setup. If BazaarLink is not offered there, exit the chat and return to hermes model.

Confirm a short response before enabling tools

Start the CLI:

hermes

Ask for a single sentence without browsing, file access or external messaging. Confirm that a response arrives, then check BazaarLink usage and Hermes diagnostics. Only after that should you enable one needed tool and test a small task in a disposable directory.

A sensible first tool test asks the agent to create and read a harmless text file, with approval required. Success means the expected tool call actually ran and returned the file content—not just that the model described what it would do. Record the Hermes version, model ID, number of model calls and result if you want a repeatable compatibility test.

Why one user task can consume several requests

An agent may call the model again after each tool result, during retries or for auxiliary work such as summarization. Hermes's provider reference notes that auxiliary models can be configured separately. Review those settings if your usage is larger than expected.

The live BazaarLink free tier currently shows base limits of 10 requests/minute and 50 requests/day, with ×1/×2 account-tier multipliers. The actual tier depends on current balance/subscription eligibility and configured thresholds. These are shared across free-model traffic. After the quota, accounts meeting paid-fallback conditions can incur normal paid charges; otherwise they are rate-limited. See the free-model rules before scheduling repeated tasks.

For illustration, if a successful task takes six calls, a 50-request allowance leaves room for eight such tasks before retries or other traffic. Measure your own task instead of borrowing that example as a forecast.

Troubleshoot the layer that failed

My recommendation is to keep the initial test deliberately small. Once a specific task works, decide whether free capacity is enough or whether a paid model is necessary. The same endpoint can simplify that switch, but it cannot make every model equally capable. For a broader quota comparison, see free LLM API options.

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API