Providers overview

Scira uses two independent provider slots: one for LLM calls (the reasoning model that runs the research agent) and one for web search (what it searches with). You configure them separately, and each has its own key.

LLM providers

Provider	Key(s)	Default model
`gateway` (default)	`AI_GATEWAY_API_KEY`	`deepseek/deepseek-v4-flash`
`xai`	`XAI_API_KEY`	`grok-build-0.1` (coding agent model)
`workers-ai`	`CLOUDFLARE_ACCOUNT_ID` + `CLOUDFLARE_API_TOKEN`	`@cf/moonshotai/kimi-k2.6`
`huggingface`	`HF_API_KEY`	`meta-llama/Llama-3.3-70B-Instruct`

Each provider page in this section covers credentials, available models, and how to configure it. Start with Vercel AI Gateway — it gives access to the widest model selection with a single key.

Local agent runtimes

Two more LLM providers run a local coding agent instead of a plain model:

Provider	Auth	Default model
`claude-code`	`claude login` session (`~/.claude`)	Claude Sonnet 4.6
`codex`	`codex login` session (`~/.codex`)	GPT-5.5

Pick them from /llm. They use your existing CLI login — no API key — and drive the AI SDK harness on your real files, with Scira's own web search and instructions. Tune effort with /thinking (Claude Code) and /reasoning (Codex). See Claude Code and Codex for setup, or run scira doctor to check each runtime is installed and logged in.

Search providers

Provider	Key	Notes
`exa` (default)	`EXA_API_KEY`	Neural search, high quality
`firecrawl`	`FIRECRAWL_API_KEY`	Crawl + search; doubles as automatic fallback
`parallel`	`PARALLEL_API_KEY`	Structured parallel search with advanced mode

FIRECRAWL_API_KEY is worth setting even if Firecrawl is not your primary provider. When Exa or Parallel returns poor results, Scira falls back to Firecrawl automatically.

Switching providers

The fastest path is the interactive wizard:

scira init

From inside the TUI:

/llm           # pick LLM provider + model
/provider      # pick search provider
/model         # change model within the current LLM provider
/key NAME val  # set a single key immediately

Or edit your config directly:

~/.scira/config.json

{
  "llmProvider": "xai",
  "model": "grok-4"
}

Changes take effect on the next run. Running /llm or /provider in the TUI applies the change immediately without restarting.

Model picker

The /model command opens an interactive picker. For Vercel AI Gateway, it fetches the live model list from the gateway and filters to models that support tool use. For xAI and Cloudflare, it calls the provider's models API and falls back to a curated static list when keys are absent.

Credential precedence

Keys are loaded in this order — first match wins:

Variables already exported in your shell
<project>/.scira/.env when running from that directory
~/.scira/.env global defaults

scira doctor   # shows which keys are present and from where
scira keys     # shows key status for the current config