Every major provider, one sign-in.
Pearl Scout (native), Codex, Claude Code, and GitHub Copilot all live behind one model picker in the chat input. Here's how each one authenticates, which models it serves, and where your keys live.
Pearl Scout (native) ↓
Pearl-account multi-provider gateway. Anthropic + OpenAI + Google + DeepSeek + PearlFibers.
Codex (OpenAI) ↓
Your OpenAI key. Frontier and reasoning models. Stored in OS keychain.
Claude Code (Anthropic) ↓
Your Anthropic key. Full Claude 4 lineup. Stored in OS keychain.
GitHub Copilot ↓
Device flow. Copilot's chat-model lineup. No key to paste.
Pearl Scout — native, multi-model.
Pearl Scout — native, multi-model.The provider built into Pearl. One sign-in, every major model your plan entitles you to. No vendor lock-in — switch models per turn without changing providers.
How it authenticates
Via your Pearl account. Sign in once with PearlFibers SSO and Scout picks up your entitlements automatically. No API key to paste. No keychain entry to manage. Account deletion at PearlFibers propagates to Pearl Scout immediately.
Models served
Pearl Scout routes to multiple underlying providers based on the model you pick:
- Anthropic — Claude 4 family (Opus, Sonnet, Haiku)
- OpenAI — GPT-class frontier and reasoning models
- Google — Gemini Pro lineup
- DeepSeek — selected frontier models
- PearlFibers — Peak AI's own models
Models surface in the picker grouped by provider, with tier icons (rocket / sparkle / zap) so the differences are visible at a glance. The exact set you see depends on your plan.
Where keys live
You don't have any provider keys to manage. Pearl Scout holds the keys; we never expose them to the desktop client.
Quota and billing
Scout requests count against your Pearl plan's monthly quota. Your dashboard at /account/usage shows current period consumption by model, by week, and by individual seat (Business+). Sales reps can raise quota mid-cycle.
Codex — your OpenAI key, your bill.
If you already pay OpenAI directly and want Pearl to route through your account.
How it authenticates
Paste your OpenAI API key (sk-…) into the provider picker once. Pearl
stores it in the OS keychain (keytar-backed) — Windows Credential Manager on Windows,
Keychain on macOS, libsecret on Linux. Pearl never sees the key in plaintext beyond
the handoff.
If keychain write fails, Pearl falls back to user preferences with an inline warning.
Models served
OpenAI's frontier (GPT-class) and reasoning models, including the latest releases as OpenAI publishes them. Pearl doesn't gate models behind plan tiers — if your OpenAI key has access, Pearl uses it.
Quota and billing
You're billed directly by OpenAI for token usage. Pearl Scout's quota doesn't apply here — your Pearl plan only covers Pearl Scout (native) requests. OpenAI's own billing dashboard is the source of truth.
When to pick this over Pearl Scout
- You already have OpenAI billing set up and want the simpler accounting.
- You want access to a brand-new OpenAI model before it lands in Pearl Scout's plan tier.
- You need OpenAI's specific quota tier (e.g. higher tokens-per-minute than the Pearl plan offers).
Claude Code — your Anthropic key.
The same model lineup you'd use in Claude's own products, signed in to Pearl with your own Anthropic API key.
How it authenticates
Paste your Anthropic API key (sk-ant-…) into the provider picker. OS
keychain storage, same as Codex.
Models served
Full Claude 4 lineup — Opus, Sonnet, Haiku — including extended-reasoning variants where Anthropic offers them. Scout's thinking blocks activate automatically when the active model supports them.
Quota and billing
Billed directly by Anthropic. Pearl's own quota doesn't apply.
Why Anthropic users specifically pick this
- Claude's extended-reasoning behavior shows up via Scout's thinking block UI.
- Anthropic's prompt-caching pricing tier applies to your key.
- Direct billing relationship with the model provider — no Pearl middle layer.
GitHub Copilot — device flow.
Use Copilot's chat models from inside Pearl. No API key to paste.
How it authenticates
GitHub device flow — Pearl shows you a code, you confirm it in a GitHub-hosted page, Pearl picks up the token from GitHub. Your Copilot subscription stays attached to your GitHub account.
Models served
Whatever chat-model lineup Copilot is currently serving — including the model picker that GitHub itself exposes inside Copilot. Pearl surfaces these alongside Scout, Codex, and Claude Code so you can compare in one session.
Quota and billing
Counts against your GitHub Copilot subscription. Pearl's quota doesn't apply.
The "use both Pearl and Copilot" case
Many teams adopt Pearl for the editor + Ask/Plan/Agent + inline-diff workflow while keeping Copilot for the completion experience they're used to. Both work in the same session — Copilot for inline completions, Scout (native) or Claude Code for chat and Agent edits.
Same workflow, every provider.
All three modes work everywhere
Ask, Plan, and Agent are agent-side abstractions — they work on every provider, not just on Scout.
Inline diff is provider-agnostic
Every Agent edit lands as an inline diff regardless of which provider generated it.
Project memory is shared
.pearl/memory/project.md is read at the start of relevant turns, no matter which provider serves the response.
Context-usage ring follows the model
Each model's context window is registered, so the ring shows real utilization whatever provider you're on.
Switching is one click
Same dropdown, same keystroke. Mid-session model changes don't lose your thread.
A Pearl account is the gate
Every provider requires a Pearl sign-in — including Codex, Claude Code, and Copilot. Sign-in is the single source of truth for entitlements, telemetry, and memory.