OpenAI

The OpenAI provider lives in crates/lime-provider-openai/ and uses the Responses API with optional WebSocket streaming.

Authentication

lime login --with-api-key                          # OpenAI is the default provider

Or via environment:

export OPENAI_API_KEY="sk-…"
export OPENAI_BASE_URL="https://api.openai.com/v1"   # optional override

lime login writes credentials under ~/.lime/; stored values override the env vars when present.

Built-in catalog

Slug	Display	Context	Max output	Capabilities
gpt-5.5	GPT-5.5	1M	90 K	Vision, parallel tools, reasoning summaries, param_preset: reasoning_and_verbosity
gpt-5.4	GPT-5.4 (default)	1M	90 K	Vision, parallel tools, reasoning summaries, param_preset: reasoning_and_verbosity
gpt-5.4-pro	GPT-5.4 Pro	1M	90 K	Vision, parallel tools, reasoning summaries, param_preset: reasoning_and_verbosity
gpt-5.4-mini	GPT-5.4 Mini	400 K	64 K	Vision, parallel tools, reasoning summaries, param_preset: reasoning_and_verbosity
gpt-5.4-nano	GPT-5.4 Nano	400 K	32 K	Vision, parallel tools, param_preset: verbosity_only
o3	o3	200 K	100 K	Vision, parallel tools, reasoning summaries, param_preset: reasoning_and_verbosity
o4-mini	o4-mini	200 K	100 K	Vision, parallel tools, reasoning summaries, param_preset: reasoning_and_verbosity
o1	o1	200 K	100 K	Vision, no parallel tools, reasoning summaries, param_preset: reasoning_only
gpt-4o	GPT-4o	128 K	16.4 K	Vision, parallel tools, param_preset: verbosity_only

Aliases

The provider accepts a small set of friendly aliases that resolve to the canonical slugs above:

Alias	Resolves to
gpt4	gpt-5.4
mini	gpt-5.4-mini
o3-mini	o3

What’s special

• Reasoning carries across turns. ZDR-encrypted reasoning items (encrypted_content) are captured, persisted in the session log, and echoed back on subsequent iterations so the model’s thinking continues across the conversation.
• WebSocket streaming. When supported, Lime upgrades to the WebSocket variant of the Responses API for lower latency.
• --reasoning-effort maps directly to the API field. Acceptable values: none, low, medium, high, xhigh.
• --service-tier forwards to the API. Use priority for priority routing on supported plans.

Common usage

lime --model gpt-5.4-mini --reasoning-effort high
lime --model o3 --service-tier priority
lime --model gpt-5.4 --output-schema ./schema.json

Custom OpenAI deployments

For Azure, an OpenAI proxy, or any drop-in compatible service that implements the Responses API, set OPENAI_BASE_URL:

export OPENAI_API_KEY="<your-key>"
export OPENAI_BASE_URL="https://my-openai-proxy.example.com/v1"

For services that implement only the Chat Completions API, use the OpenAI-compatible provider instead.