OpenAI (direct)

Connect Toby directly to the OpenAI API. Personas that use the openai provider send requests to OpenAI with bare model ids (no provider/ prefix).

Get an API key

1. Sign in at platform.openai.com.
2. Open API keys (under your organization settings).
3. Create a new secret key. Copy it immediately—OpenAI shows the full value only once.
4. Ensure your account has billing enabled and usage limits that fit your expected volume.

Store the key in Toby, not in shell profiles or committed files.

Configure in Toby

toby config

1. Go to AI → OpenAI → API Token.
2. Paste your API key and save.

Toby writes the token to ~/.toby/credentials.json.

For a persona, set AI Provider to openai and choose a model from the list (or type a supported model id if you know it is available on your account).

Recommended models

These models are built into Toby’s configure picker and work well for everyday use. All use direct ids (examples below).

Everyday chat (recommended default)

Model	Role
gpt-5-mini	Default for the built-in Toby persona. Strong balance of quality, speed, and cost for chat, summarization, and organization.
gpt-4.1-mini	Reliable fallback for planning and other auxiliary calls.
gpt-4.1-nano	Default pretreatment model when using OpenAI direct (fast classification).

Faster or lower cost

Model	Role
gpt-5-nano	Lighter GPT-5 tier for quick triage or high-volume schedules.
gpt-4.1-nano	Cheapest GPT-4.1 family option for simple classification-style work.
gpt-4o-mini	Older but still capable small model if your account still lists it.

Higher quality or reasoning

Model	Role
gpt-5	More capable GPT-5 tier when answers need extra depth.
gpt-4.1	Strong general model when you want quality without the GPT-5 line.
gpt-4o	Previous flagship multimodal model; still useful if you rely on it today.
o3	Reasoning-focused work (complex planning, multi-step analysis).
o4-mini	Lighter reasoning model when you want some “think longer” behavior at lower cost than o3.

Start with gpt-5-mini for most personas. Move to gpt-5 or o3 when you notice quality limits; use gpt-5-nano for cron-style schedules that run often.

Prompt caching

Toby sends a stable OpenAI promptCacheKey on chat turns so repeated system prompts can hit OpenAI prompt caching when your model and account support it.

Environment variables

Variable	Purpose
TOBY_PRETREAT_MODEL	Override the pretreatment model (bare OpenAI id, e.g. gpt-4.1-nano)
TOBY_PRETREAT_DELTA	Set to 0 to disable follow-up delta pretreatment (default: enabled)

OpenAI credentials are not read from OPENAI_API_KEY by Toby today—use configure or ~/.toby/credentials.json via backup/restore.

See also

• Vercel AI Gateway — one key, many vendors
• AI providers overview
• Personas