Documentation / Getting started
Configuration
Choose a provider and a model, set API keys, and enable the optional tools.
Seshat needs at least one model provider. You can set it up from the command line, from the terminal interface, or with environment variables.
From the command line
seshat config --provider anthropic --api-key sk-ant-... # set a provider and its key
seshat config --model anthropic:claude-sonnet-5-5 # choose the model (provider:model)
seshat config --print # show the current configuration
seshat config also takes --base-url for a custom endpoint, --cwd for the default working directory, --db for the database path, and --search to set only the web search keys. The keys you save this way are stored encrypted, per provider, in the local database.
In the terminal interface, press ctrl+, for providers and ctrl+m for models (the models of the providers you have configured). A key entered in the providers screen is stored the same way, so it never reaches seshat.json: that file holds settings, not secrets. The key that protects the database is secret.key, next to it in the runtime folder.
With environment variables
Set the variable of the provider you want. This is the simplest way for servers and scripts:
export ANTHROPIC_API_KEY=sk-ant-...
Other settings follow a simple ladder:
The main variables are SESHAT_MODEL (provider:model), SESHAT_API_KEY, SESHAT_RUNTIME_ROOT, SESHAT_DB_PATH, SESHAT_CWD, SESHAT_DEBUG, and for the cloud providers SESHAT_PROVIDER_BASE_URL, SESHAT_PROVIDER_REGION, SESHAT_PROVIDER_PROJECT_ID and SESHAT_PROVIDER_RESOURCE.
Supported providers
| Provider | ID | Authentication | Setting |
|---|---|---|---|
| Anthropic | anthropic | API key | ANTHROPIC_API_KEY |
| OpenAI | openai | API key | OPENAI_API_KEY |
| Mistral AI | mistral | API key | MISTRAL_API_KEY |
| Google Gemini | gemini | API key | GOOGLE_API_KEY or GOOGLE_GEMINI_API_KEY |
| DeepSeek | deepseek | API key | DEEPSEEK_API_KEY |
| Ollama | ollama | None | Local, localhost:11434 |
| OpenRouter | openrouter | API key | OPENROUTER_API_KEY |
| Codex | codex | OAuth (a ChatGPT account) | Choose Codex in the providers of the interface for a code and the browser, or seshat login, or CODEX_API_KEY or CODEX_ACCESS_TOKEN |
| Z.ai | z-ai | API key | Z_AI_API_KEY or ZHIPUAI_API_KEY |
| MiniMax | minimax | API key | MINIMAX_API_KEY |
| Cloudflare Workers AI | cloudflare-workers-ai | API key | CLOUDFLARE_API_KEY and CLOUDFLARE_ACCOUNT_ID |
| OpenCode Zen | opencode | API key | OPENCODE_API_KEY |
| Kimi (Moonshot AI) | kimi | API key | MOONSHOT_API_KEY or KIMI_API_KEY |
The models each provider offers, with their context sizes and capabilities, are in Providers and models.
About 190 more providers work without any setting of their own: name one in a model string (groq:..., together:..., fireworks-ai:...), set the key variable it uses, and Seshat calls it with the generic OpenAI-compatible client. They come from the public models.dev catalog; see Providers and authentication.
Optional tools
Some tools exist in every installation but do nothing until you give them what they need.
| Tool | What it does | What it needs |
|---|---|---|
generate_image | Image generation | An image_generation provider (openai or gemini) in the configuration, and a key. It reuses the key of your model provider when they are the same |
text_to_speech, speech_to_text | Speech synthesis and transcription | A text_to_speech or speech_to_text provider in the configuration, and a key |
web_search | Search the web | A key for one search service: TAVILY_API_KEY, EXA_API_KEY, JINA_API_KEY or LANGSEARCH_API_KEY, or SEARXNG_BASE_URL. seshat config --search sets them. WEB_SEARCH_PROVIDER picks the service |
devto_publish | Publish to DEV | DEV_TO_API_KEY |
notebook_* | Work in a live Jupyter kernel | JUPYTER_SERVER_URL (default http://localhost:8888) and JUPYTER_TOKEN |
Reference
Updated on 2026-10-13