log inget an API key

Hermes Agent with SayGM

Nous Research's self-improving agent, running on SayGM's models.

Model

Claude Opus 5.5Best quality

In $4.00$3.57Out $20.00$17.86/1M

Sign in to fill these snippets with a SayGM key for Hermes Agent.

Sign in

Setup for Claude models

Use this setup with the models listed under Claude below.

  1. 1~/.hermes/config.yaml
    yaml
    providers:
  saygm-claude:
    base_url: https://api.saygm.com
    key_env: SAYGM_API_KEY
    api_mode: anthropic_messages
    models:
      - claude-opus-5-5
      - claude-sonnet-5

model:
  provider: custom:saygm-claude
  default: claude-opus-5-5
  2. 2~/.hermes/.env
    bash
    SAYGM_API_KEY=<YOUR_SAYGM_KEY>
  3. 3Run
    bash
    hermes

Switch models with /model, which lists the provider's models, or name one: /model custom:saygm-claude:claude-sonnet-5.

First request

  1. 1Send a first request
    bash
    hermes -z "say gm"

Setup for GPT and open models

Use this setup with the models listed under GPT, Open models and Private (TEE) below.

  1. 1~/.hermes/config.yaml
    yaml
    providers:
  saygm:
    base_url: https://api.saygm.com/v1
    key_env: SAYGM_API_KEY
    api_mode: chat_completions
    models:
      - gpt-6-luna
      - gpt-6-astra
      - kimi-k3-tee

model:
  provider: custom:saygm
  default: gpt-6-luna
  2. 2~/.hermes/.env
    bash
    SAYGM_API_KEY=<YOUR_SAYGM_KEY>
  3. 3Run
    bash
    hermes

Switch models with /model, which lists the provider's models, or name one: /model custom:saygm:gpt-6-astra.

First request

  1. 1Send a first request
    bash
    hermes -z "say gm"

Troubleshooting

Adding another model.
Add its id under models in the provider, then switch to it with /model.
Hermes' docs spell the keys api and transport.
Hermes accepts both spellings. base_url and api_mode match its current example config.
Hermes picks the wrong context length.
Add context_length under the provider with the model's real limit.
401 Unauthorized.
Hermes reads SAYGM_API_KEY from ~/.hermes/.env. Check the variable name matches key_env.

Models you can use with Hermes Agent

Hermes runs SayGM's Claude, GPT and open models, with one provider for Claude and one for the rest.

Claude

9 models
  • Claude Opus 5.5claude-opus-5-5Best qualityThe newest Opus: hard coding and long agentic work at a lower Opus price.In $4.00$3.57Out $20.00$17.86/1M
  • Claude Sonnet 5claude-sonnet-5Best valueEveryday coding and agent work at a fraction of Opus pricing.In $2.00$1.79Out $10.00$8.93/1M
  • Claude Fable 5.1claude-fable-5-1Anthropic's most capable model, for the most demanding long-horizon tasks.In $10.00$8.93Out $50.00$44.65/1M
  • Claude Fable 5claude-fable-5Anthropic's top tier before Fable 5.1, for deep reasoning and agentic work.In $10.00$8.93Out $50.00$44.65/1M
  • Claude Opus 5claude-opus-5Opus 5 for complex coding and agents that need careful reasoning.In $5.00$4.47Out $25.00$22.33/1M
  • Claude Opus 4.8claude-opus-4-8Opus 4.8 for teams whose prompts are tuned to the Opus 4 line.In $5.00$4.47Out $25.00$22.33/1M
  • Claude Opus 4.7claude-opus-4-7Opus 4.7 for workflows pinned to that release.In $5.00$4.47Out $25.00$22.33/1M
  • Claude Sonnet 4.6claude-sonnet-4-6Sonnet 4.6 for workflows pinned to the previous Sonnet.In $3.00$2.68Out $15.00$13.40/1M
  • Claude Haiku 4.5claude-haiku-4-5Fast, low-cost Claude for quick edits, subagents and high-volume calls.In $1.00$0.89Out $5.00$4.47/1M

GPT

12 models
  • GPT-6 Astragpt-6-astraBest qualityOpenAI's top GPT-6 model, for the hardest coding and reasoning work.In $10.00$8.15Out $50.00$40.75/1M
  • GPT-6 Lunagpt-6-lunaBest valueLow-cost GPT-6 for fast coding loops and high-volume agent turns.In $0.10$0.08Out $0.50$0.41/1M
  • GPT-6 Solgpt-6-solMid-priced GPT-6 for day-to-day coding with strong reasoning.In $2.00$1.63Out $10.00$8.15/1M
  • GPT-5.6 Solgpt-5.6-solThe larger GPT-5.6 model, for demanding coding and analysis.In $4.00$3.26Out $20.00$16.30/1M
  • GPT-5.6 Terragpt-5.6-terraMid-sized GPT-5.6 balancing capability and cost.In $2.00$1.63Out $12.00$9.78/1M
  • GPT-5.6 Lunagpt-5.6-lunaLow-cost GPT-5.6 for quick, high-volume requests.In $0.20$0.16Out $1.20$0.98/1M
  • GPT-5.5gpt-5.5GPT-5.5 reasoning model for workflows tuned to it.In $5.00$4.08Out $30.00$24.45/1M
  • GPT-5.4gpt-5.4GPT-5.4 for general coding and writing at a moderate price.In $2.50$2.04Out $15.00$12.23/1M
  • GPT-5.4 minigpt-5.4-miniSmall GPT-5.4 for fast, inexpensive everyday tasks.In $0.75$0.61Out $4.50$3.67/1M
  • GPT-5.4 nanogpt-5.4-nanoThe smallest GPT-5.4, for classification and simple high-volume calls.In $0.20$0.16Out $1.25$1.02/1M
  • o3OpenAI's o3 reasoning model for multi-step math, science and code.In $2.00$1.63Out $8.00$6.52/1M
  • o4-miniCompact o-series reasoning at a low price.In $1.10$0.90Out $4.40$3.59/1M

Open models

11 models
  • gpt-oss-20bOpenAI's small open-weight model for cheap, simple tasks.In $0.07Out $0.30$0.28/1M
  • Kimi K3kimi-k3Kimi K3, an open-weight reasoning model built for coding agents.In $3.00$1.02Out $15.00$5.09/1M
  • GLM-5.3glm-5.3Z.ai's GLM-5.3 open-weight model for coding and agent tasks.In $1.40$0.81Out $4.40$2.54/1M
  • GLM-5.3-Flashglm-5.3-flashThe fast, low-cost GLM-5.3 for quick coding help.In $0.15$0.13Out $0.50$0.45/1M
  • GLM-5.2glm-5.2GLM-5.2 open weights for coding and reasoning.In $1.40$0.44Out $4.40$1.38/1M
  • DeepSeek-V4.1-Flashdeepseek-v4.1-flashDeepSeek's fast V4.1 model for low-cost coding and chat.In $0.30$0.09Out $1.20$0.36/1M
  • DeepSeek-V4-Flash-0731deepseek-v4-flash-0731DeepSeek V4 Flash for inexpensive, quick coding help.In $0.44$0.13Out $1.32$0.39/1M
  • Qwen3.8-27Bqwen3.8-27bQwen3.8 27B, a compact open-weight model for everyday coding.In $0.50$0.05Out $3.00$0.32/1M
  • Qwen3.6-35B-A3Bqwen3.6-35b-a3bQwen3.6 35B mixture-of-experts, fast and cheap for light tasks.In $0.248$0.052Out $1.485$0.310/1M
  • MiMo-V2.6-Pro-UltraSpeedmimo-v2.6-pro-ultraspeedXiaomi MiMo V2.6 Pro tuned for speed, with a 1M-token context.In $4.35$3.92Out $8.70$7.83/1M
  • Ornith 1.5 397Bornith-1.5-397bOrnith 1.5 397B, a large open-weight reasoning model.In $1.40$0.70Out $4.40$2.20/1M

Private (TEE)

13 models
  • Kimi-K3 (TEE)kimi-k3-teePrivate (TEE)Kimi K3, an open-weight coding and agent model, run inside a TEE.In $3.00$1.03Out $15.00$5.17/1M
  • Kimi-K2.6 (TEE)kimi-k2.6-teeKimi K2.6 open weights inside a TEE, for private agent work.In $0.50$0.17Out $2.85$0.98/1M
  • GLM-5.2 (TEE)glm-5.2-teeGLM-5.2 run inside a TEE, for coding on private code.In $1.40$0.44Out $4.40$1.38/1M
  • GLM-5.1 (TEE)glm-5.1-teeGLM-5.1 run inside a TEE, for private reasoning tasks.In $1.40$0.48Out $4.40$1.52/1M
  • DeepSeek-V4-Flash-0731 (TEE)deepseek-v4-flash-0731-teeDeepSeek V4 Flash inside a TEE, for low-cost private work.In $0.44$0.22Out $1.32$0.66/1M
  • DeepSeek-V3.2 (TEE)deepseek-v3.2-teeDeepSeek V3.2 inside a TEE, for private general-purpose tasks.In $1.00$0.48Out $1.00$0.48/1M
  • Qwen3.6-27B (TEE)qwen3.6-27b-teeQwen3.6 27B inside a TEE, for private everyday tasks.In $0.30$0.15Out $2.00$0.98/1M
  • Qwen3.5-397B-A17B (TEE)qwen3.5-397b-a17b-teeQwen3.5 397B, a large open-weight model run inside a TEE.In $0.45$0.16Out $3.00$1.03/1M
  • Qwen3-235B-A22B-Thinking-2507 (TEE)qwen3-235b-a22b-thinking-2507-teeQwen3 235B thinking model inside a TEE, for private step-by-step reasoning.In $0.2989$0.1449Out $1.1957$0.5797/1M
  • Qwen3-32B (TEE)qwen3-32b-teeQwen3 32B inside a TEE, for low-cost private requests.In $0.104$0.051Out $0.416$0.204/1M
  • Gemma 4 31B turbo (TEE)gemma-4-31b-turbo-teeGoogle's Gemma 4 31B open weights inside a TEE, low cost and private.In $0.12$0.06Out $0.37$0.18/1M
  • Mistral Nemo (TEE)mistral-nemo-instruct-2407-teeMistral Nemo inside a TEE, a very low-cost model for simple private tasks.In $0.0245$0.0120Out $0.0978$0.0479/1M
  • Nemotron 3 Nano Omni 30B (TEE)nemotron-3-nano-omni-30b-teeNVIDIA Nemotron 3 Nano Omni inside a TEE, very low cost and private.In $0.0245$0.0110Out $0.0978$0.0439/1M

About Hermes Agent

Hermes Agent is Nous Research's open-source, self-improving agent. It runs in your terminal or on a server you reach from Telegram, Discord, Slack and other chat apps, turns what it learns into reusable skills, and keeps memory across sessions. People use it as a long-running personal agent for scheduled jobs and everyday tasks, on any model provider.