Hermes Agent Web Search: Use a BazaarLink Model with :online
Point Hermes Agent at BazaarLink as a custom endpoint and add :online to the model name for live web results with citations. Config, switching and billing.
Hermes Agent can talk to any OpenAI-compatible endpoint. Set BazaarLink as its model provider, add :online to the model name, and each request is searched first and answered with citations. Models and search share one API key and one bill.
Step 1: add BazaarLink as a custom endpoint
~/.hermes/.env:
BAZAARLINK_API_KEY=your BazaarLink API key
~/.hermes/config.yaml:
providers:
bazaarlink:
api: https://api.bazaarlink.ai/v1
key_env: BAZAARLINK_API_KEY
default_model: openai/gpt-6-luna
transport: chat_completions
model:
provider: custom:bazaarlink
default: openai/gpt-6-luna
Or run hermes model and choose "Custom endpoint".
Step 2: switch to :online when you need fresh information
/model custom:bazaarlink:openai/gpt-6-luna:online
Switch back to openai/gpt-6-luna when you do not need search.
Why switch instead of leaving it on
:online searches once per request and search is billed separately from tokens. An agent calls the model many times per task, so leaving :online on accumulates search charges. Use the plain model for reasoning and coding, and :online only when you need current data.
Tuning and verifying
Depth, result count, time range, country and language can be set with plugins or web_search_options (see adding web search with :online). To verify, ask something time-sensitive; the answer should include source links and one search should appear in your usage.
Billing
Per-search billing per the price table. Failed searches are not charged.
FAQ
Hermes already has a Tavily backend; why this? That backend needs its own Tavily key. :online keeps models and search on one account and one bill, and the two can coexist. Model names follow the BazaarLink catalog.
TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API