How do I use DeepSeek in Hermes Agent?
Add a standardcompute entry under providers: in ~/.hermes/config.yaml (api https://api.stdcmpt.com/v1, key_env STANDARDCOMPUTE_API_KEY) and set model.provider to custom:standardcompute with default deepseek/deepseek-v4.1-flash: Hermes then runs on DeepSeek V4.1 Flash on a plan from $19 a month, and new requests get HTTP 402 when the compute budget reaches 100%.
# DeepSeek V4.1 Flash by default, V4 Pro one /model away
# Put STANDARDCOMPUTE_API_KEY=YOUR_API_KEY in ~/.hermes/.env
providers:
standardcompute:
api: https://api.stdcmpt.com/v1
key_env: STANDARDCOMPUTE_API_KEY
transport: chat_completions
default_model: deepseek/deepseek-v4.1-flash
models:
deepseek/deepseek-v4.1-flash: { context_length: 400000, supports_vision: true }
deepseek/deepseek-v4-pro: { context_length: 400000 }
model:
provider: custom:standardcompute
default: deepseek/deepseek-v4.1-flash
Hermes reads ~/.hermes/.env at start, so a key exported only in another shell is not seen. context_length is 400,000 so Hermes compresses before turns get costly.
Why the config names supports_vision
Hermes decides per model whether to send pixels or route an image through its vision_analyze helper. Hermes' own docs list V4.1 Flash as vision-capable and V4 Pro as text-only; for a custom provider, supports_vision: true on the Flash entry tells it so. On V4 Pro, images go to the auxiliary vision model, which by default skips a text-only main model.
What works with DeepSeek in Hermes
| Hermes feature | On DeepSeek through Standard Compute |
|---|---|
| Reasoning | Hermes sends reasoning_effort medium unless you change it with /reasoning; that overrides the Models page setting |
| Context | Set by context_length. Hermes needs at least 64K and compresses as the session nears the number you give it |
| Switching mid-session | /model custom:standardcompute:deepseek/deepseek-v4-pro |
| Titles and compression | Use the main model unless you set auxiliary models (hermes model, then Configure auxiliary models) |
From Hermes Agent's providers, configuration and vision docs and its source, checked October 14, 2026.
What a Hermes day on DeepSeek costs
Hermes runs scheduled jobs and messaging replies when you are not at the keyboard, so a quiet hour can still use tokens.
| Per day | V4.1 Flash, peak | Off-peak |
|---|---|---|
| Typical day | $0.42 | $0.21 |
| Busy day | $4.31 | $2.16 |
DeepSeek list prices per million tokens for V4.1 Flash: $0.30 input, $0.006 cache hit, $1.20 output at peak, half off-peak. Source. The Agent Burn Index measures coding agents such as OpenCode, Cline and Kilo; Hermes Agent is not broken out, so these figures use its median and 90th-percentile coding-agent day (September 13 to October 12, 2026) as a stand-in. Priced at the list rates above; your own days will differ.
The catch
Hermes has a built-in DeepSeek provider (DEEPSEEK_API_KEY, models deepseek-flash and deepseek-v4-pro in the hermes model picker). Scheduled jobs that run overnight can land in DeepSeek's off-peak hours, when DeepSeek's own price halves.
Where Standard Compute fits
A monthly cap for an agent that runs while you sleep: when the budget is used, requests stop with HTTP 402 instead of running up a bill. The same key serves Claude or GPT when a Hermes job needs more than DeepSeek.
FAQ
Do I still need OPENAI_BASE_URL for Hermes?
No. Hermes no longer reads OPENAI_BASE_URL for custom endpoints, and LLM_MODEL was removed. Older guides use both; the providers: block replaces them.
Why do I get a 401 with key_env set?
If the variable named in key_env is empty, Hermes sends a placeholder key. Put STANDARDCOMPUTE_API_KEY=YOUR_API_KEY in ~/.hermes/.env and run hermes doctor.
Can I use the standardcompute launcher instead?
Yes: standardcompute launch hermes opens Hermes in a separate profile on smart routing. The config on this page is for pinning DeepSeek in your own setup.