Hermes Agent on a security model that answers

Add this API to Hermes Agent as one providers entry in ~/.hermes/config.yaml: the api URL, a key_env name, the chat_completions transport and your model ids. The key stays in .env, /model switches builds mid-session, and the page prices an agent turn and decodes each error.

Steps checked November 6, 2026 · prices as of November 18, 2026

Hermes Agent will drive a shell, read files and spawn subagents, which is exactly what you want when triaging an authorized target, right up until the stock model behind it refuses. Hermes registers each OpenAI-compatible endpoint as a provider of its own under providers in ~/.hermes/config.yaml. With that one entry a security-tuned or abliterated build runs the agent, tool calls included, billed per token. The same ids also work in OpenClaw and OpenCode.

What you need

Hermes Agent installed, a key from the console, and two files under ~/.hermes/: config.yaml for settings and .env for secrets. Precedence runs CLI flags, then config.yaml, then .env, then built-in defaults. A providers entry refers to its key by variable name through key_env, which keeps the key out of config.yaml entirely. Older write-ups use a custom_providers list; Hermes still accepts it and hermes update converts it, but new setups use the mapping below. To try a security build first, the home page prompt box gives a few free turns and the playground runs the same catalog.

Shortest path to a first reply

  1. Make a key just for Hermes, so its requests are easy to separate on the usage page when you bill an engagement, and write it into the env file as one line:
.env
1# ~/.hermes/.env
2PENTESTLLM_API_KEY=sk-ptl-live-…
  1. Add the entry to config.yaml. The ids under models are a first set; every id on the models page fits the same mapping:
config.yaml
1# ~/.hermes/config.yaml
2model:
3 default: glm-5.3-uncensored
4 provider: custom:pentestllm
5
6providers:
7 pentestllm:
8 api: https://api.pentestllm.ai/v1
9 key_env: PENTESTLLM_API_KEY
10 transport: chat_completions
11 default_model: glm-5.3-uncensored
12 models:
13 glm-5.3-uncensored: {}
14 glm-5.3-flash-uncensored: {}
15 deepseek-v4.1-flash-abliterated: {}
16 deepseek-v4-pro-uncensored: {}
17 glm-5.3-abliterated: {}
18 kimi-k3.1-pro-uncensored: {}
  1. Start hermes and give it something in scope, such as turning a raw port scan into a host table. The request reaches the console’s usage page within seconds, token counts and charge included.

Prefer the shell? The same configuration as commands. A dotted path writes to config.yaml; an UPPER_SNAKE name writes to .env:

shell
1hermes config set PENTESTLLM_API_KEY sk-ptl-live-…
2hermes config set providers.pentestllm.api https://api.pentestllm.ai/v1
3hermes config set providers.pentestllm.key_env PENTESTLLM_API_KEY
4hermes config set providers.pentestllm.transport chat_completions
5hermes config set model.provider custom:pentestllm
6hermes config set model.default glm-5.3-uncensored

Change builds during a session with /model. The switch holds for that session only, unless you add --global, which saves it to config.yaml:

hermes
1/model custom:pentestllm:glm-5.3-flash-uncensored
2/model custom:pentestllm:deepseek-v4.1-flash-abliterated --global

Each field of the entry

api
https://api.pentestllm.ai/v1
the endpoint including /v1; Hermes also accepts base_url for this field
key_env
PENTESTLLM_API_KEY
the .env variable holding the full sk-ptl- key
transport
chat_completions
the OpenAI chat format; every build in the catalog speaks it
default_model
glm-5.3-uncensored
the id this provider opens with
models
id: {}
per-id settings; put context_length beneath an id to override the value Hermes detects
model.provider
custom:pentestllm
picks this entry; model.default names the id Hermes launches with

Which build to start with

Every file read, shell command and subagent in Hermes goes through a tool call. Before you hand a new id a long job, give it one small task and confirm a tool actually fires. Cost tracks the input rate, since each turn resends the conversation plus every tool result so far. The cost column below prices one turn of 30,000 tokens in and 1,000 out from the catalog as of 2026-11-18.

BuildSingle turnSession of 30 turns
GLM-5.3 Uncensored$0.0093$0.28
GLM-5.3 Flash Uncensored$0.0035$0.11
DeepSeek V4.1 Flash Abliterated$0.03$0.81
DeepSeek V4 Pro Uncensored$0.03$0.93
GLM-5.3 Abliterated$0.07$2.10
Kimi K3.1 Pro Uncensored$0.04$1.22
Uncensored models and their prices per million tokens
GLM-5.3 Uncensoredglm-5.3-uncensoredUncensored$0.282$0.846
GLM-5.3 Flash Uncensoredglm-5.3-flash-uncensoredUncensored$0.1053$0.3685
DeepSeek V4.1 Flash Abliterateddeepseek-v4.1-flash-abliteratedAbliterated$0.8488$1.6975
DeepSeek V4 Pro Uncensoreddeepseek-v4-pro-uncensoredUncensored$0.97$1.94
GLM-5.3 Abliteratedglm-5.3-abliteratedAbliterated$2.1825$4.365
Kimi K3.1 Pro Uncensoredkimi-k3.1-pro-uncensoredUncensored$1.164$5.82

Prices per million tokens, read from the live catalog on 2026-11-18.

If the agent’s job is mostly reviewing code for vulnerabilities, the coding guide compares the code builds and prices a full day on each. The AI red teaming page covers what an agent should and should not do during an exercise.

Confirm it works

hermes status prints the current model, the active provider and which providers hold a credential, without revealing any key. From that same shell, the key should return the model list:

shell
1hermes status
2curl https://api.pentestllm.ai/v1/models -H "Authorization: Bearer $PENTESTLLM_API_KEY"

Then fire one request shaped like Hermes’. The -i flag shows the response headers, which carry a request id and, on a non-streamed reply, the charge in USD.

shell
1curl -i https://api.pentestllm.ai/v1/chat/completions \
2 -H "Authorization: Bearer $PENTESTLLM_API_KEY" \
3 -H "Content-Type: application/json" \
4 -d '{"model": "glm-5.3-uncensored", "max_tokens": 16, "messages": [{"role": "user", "content": "Reply with the single word ok."}]}'

When something fails

  • 401 invalid_api_key: bad key, revoked key, or a key that never made it into the request. Look for the line in ~/.hermes/.env, make sure key_env names that exact variable, and copy a fresh key from the console if needed.
  • 402 insufficient_balance: the balance cannot cover this call; add credit in the console. A 402 monthly_limit_reached instead means the key reached the monthly spend ceiling you gave it.
  • 404 model_not_found: model.default holds an id that is not on the models page. Paste the id from that page into hermes config set model.default <id>.
  • 429 rate_limited: the key is over its rate limit. Read retry-after on the response.
  • 400 on the very first run: the Hermes FAQ attributes this to a model name that does not match. Compare the output of hermes config show, field model.default, with the models page.
  • Hermes ignores every custom endpoint: a leftover custom_providers key in config.yaml is no longer read as a YAML list. Move the entry into the providers mapping shown here and run hermes doctor.

Sources

Field names come from Hermes Agent’s providers page and its configuration guide, read 2026-11-06. Prices come from the live catalog when the page renders.

Questions

Yes. Hermes treats each OpenAI-compatible endpoint as its own named provider in ~/.hermes/config.yaml, declared with api, key_env and transport chat_completions. The entry on this page targets https://api.pentestllm.ai/v1, after which every build on the models page is reachable through it.

Either set model.default to the id and model.provider to custom:pentestllm, or type /model custom:pentestllm:<id> during a session. Append --global to that /model command to persist the choice in config.yaml. Paste the id from the models page instead of typing it from memory.

In ~/.hermes/.env, on a single PENTESTLLM_API_KEY=... line. hermes config set with an UPPER_SNAKE name lands in the same file. The providers entry picks it up via key_env, so config.yaml never holds the key.

You pay per token at the model’s listed price out of prepaid credit, with no subscription required. The table above prices a typical agent turn per build at the current rate. When a session repeats the same prefix, the cached input tokens are charged at that model’s discounted cache price.

Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.

Get a keyPrepaid from $10, no subscription.

New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.

Help