sSara

Coding agents

Point Claude Code or Codex at Sara and run agentic coding on your own metered keys.

Claude Code

Claude Code speaks the Anthropic Messages API. Set these environment variables and launch claude:

ANTHROPIC_BASE_URL=https://api.sara-ai.kz \
ANTHROPIC_API_KEY=$SARA_API_KEY \
ANTHROPIC_AUTH_TOKEN=$SARA_API_KEY \
ANTHROPIC_DEFAULT_OPUS_MODEL=gemma-4-31b \
ANTHROPIC_DEFAULT_SONNET_MODEL=gemma-4-31b \
ANTHROPIC_DEFAULT_HAIKU_MODEL=gemma-4-31b \
claude

Codex

Codex speaks the OpenAI Responses API. Configure ~/.codex/config.toml:

model = "gemma-4-31b"
model_provider = "sara"

[model_providers.sara]
name = "Sara"
env_key = "SARA_API_KEY"
base_url = "https://api.sara-ai.kz/v1"
wire_api = "responses"

Use a virtual Sara key — usage is metered and rate-limited per key like any other request.