Documentation

Getting Started

Get up and running with WayMore Chat in a few simple steps. Use the hosted chat or integrate via REST.

WayMore CLI

Your terminal coding agent — powered by WayMore AI. Code, debug, and ship from the command line.

Installation

Install with a single command. The installer downloads the latest binary and configures your shell automatically.

Verify Installation

Open a new terminal window after installing, then run:

Signing in

Run waymore login. The CLI prints a short code and a URL, and waits.

Open chat.waymore.ai/cli-auth in a browser, sign in if you are not already, check that the device shown matches the machine you are sitting at, and approve it. The CLI picks up the approval within a few seconds and stores its credentials — there is nothing to copy and paste.

The code is valid for 10 minutes. If it expires, run waymore login again.

Approving creates an API key named WayMore CLI — <your device name>, visible alongside your other keys in the dashboard. Revoking that key signs the device out.

Devices

Each sign-in you approve creates its own key, and each active key uses one device slot on your plan: 3 on Free and Lite, up to 20 on Ultimate — the full table is in Billing & Plans. WayMore Vibe and its Android app use the same slots.

A slot belongs to the key, not the machine. Signing in again on the same machine does not free the old slot, and if the CLI and Vibe on one laptop each ask you to approve a sign-in, that laptop uses two. Revoking a key frees its slot.

Approving a sign-in while at the limit fails with device_limit_reached; revoke a key you no longer use, or move up a tier.

An administrator can also block an individual device, which fails with device_blocked. That one survives re-pairing — signing in again will not clear it, so contact whoever administers your account.

Direct downloads

The installer above auto-detects your platform and architecture, downloads the latest build, verifies its SHA256 and installs to ~/.local/bin — it is the recommended route. If you need the binary directly (air-gapped machines, custom provisioning, container images), these links always serve the current release:

Downloading directly skips the checksum verification the installer performs, so verify against the release manifest if integrity matters for your use case. On macOS and Linux you will also need to mark the binary executable (chmod +x).

Prefer a desktop app? See WayMore Vibe.

Verifying a download

The release manifest names the current version and, for each platform, the exact URL, SHA256 and byte size of that release:

Platform keys are linux-x86_64, linux-aarch64, darwin-aarch64, darwin-x86_64, windows-x86_64 and windows-aarch64.

Pinning a version

The -latest links move with every release. For reproducible builds — container images, provisioning scripts, locked-down environments — use the version-pinned form instead, which never changes once published:

Read latest_version from the manifest to find the current release number.

Updating

Re-run the same install command to upgrade in place. The direct links above are unversioned and always point at the current release, so re-downloading updates too.

Quick Start

With the CLI installed and signed in, run it inside a project directory:

WayMore Vibe

The desktop coding agent — the same agent as the WayMore CLI, in a graphical app rather than a terminal.

WayMore Vibe is the desktop application for WayMore AI — the same coding agent as the WayMore CLI, in a graphical app rather than a terminal.

Vibe is included in every subscription plan, alongside the CLI and WayMore Chat.

Download

Download WayMore Vibe for your system — this shared link detects your operating system and sends you to the matching installer. If your Mac's processor cannot be detected, you can choose Apple Silicon or Intel.

Visit the WayMore Vibe download page to explore Chat, Corporate, and Code, compare plans, and learn about cloud and on-premises deployments. Direct downloads remain below for scripts and version-pinned installations.

All links are stable "always-latest" URLs — they always serve the current release, so you can bookmark them or reuse them in provisioning scripts.

Windows

Not sure which you need? Open Settings → System → About and check System type: x64-based processor means the x86_64 build, ARM-based processor means the arm64 one.

macOS

If you are unsure which Mac you have, click the Apple menu → About This Mac. A "Chip" entry starting with Apple means Apple Silicon; a "Processor" entry naming Intel means Intel.

Linux

Not sure which you need? Run uname -m — x86_64 means the x86_64 build, aarch64 (or arm64) means the arm64 one.

The AppImage needs no installation — mark it executable and run it:

chmod +x WayMore-Vibe-latest-x86_64.AppImage
./WayMore-Vibe-latest-x86_64.AppImage

Linux system requirements

Vibe needs glibc 2.34 or newer, the same on both architectures, plus GTK 3 and WebKit2GTK 4.1. Check yours with ldd --version.

Ubuntu 20.04, Debian 11, RHEL 8 and openSUSE Leap 15.x ship glibc 2.28–2.31 and are too old to run Vibe.

The .deb and .rpm declare this requirement, so on a system that is too old apt or dnf refuses the install and tells you why.

The AppImage carries no such metadata. It will download and run on any distribution, and on one below 2.34 it exits immediately with a message about GLIBC_2.34 not being found — that is this requirement, not a corrupt download.

On a distribution that is too old, use the WayMore CLI instead: it needs only glibc 2.28, on both architectures.

Android

The Android build is distributed as a sideloadable APK rather than through the Play Store, and it is signed with a debug key — Android will warn that it comes from an unknown developer, and you have to allow installs from whichever app you downloaded it with. It has no auto-updater: download the APK again to upgrade.

Android system requirements

Requires Android 6.0 (API 23) or newer. Unlike the desktop downloads there is no architecture to choose: the APK carries all four (arm64-v8a, armeabi-v7a, x86, x86_64), so it installs on any phone, tablet or emulator.

It asks for two permissions — INTERNET and ACCESS_NETWORK_STATE — plus one it defines for its own components. All three are granted at install; Android never prompts for them. It requests no access to your camera, microphone, location, contacts, photos or files.

Signing in

Vibe uses your WayMore account and the same browser approval as the WayMore CLI:

1. Open Vibe and follow its prompt to sign in.
2. An approval page opens in your browser at chat.waymore.ai/cli-auth. Sign in if you are not already, check that the device shown is yours, and approve it.
3. Vibe picks up the approval within a few seconds.

The code is valid for 10 minutes. If it expires, start again from Vibe.

The approval page is written for the CLI, even when the request comes from Vibe: it is titled "Authorize WayMore CLI", and the new key is named WayMore CLI — <your device name>. Revoking that key signs Vibe out on that device.

The desktop app runs its engine on your machine. The Android app connects directly to WayMore Cloud, and a desktop app can be switched to it. Each of these connections has its own sign-in.

Devices and requests

Vibe shares your plan with the CLI:

• Devices: every sign-in you approve uses one device slot, including the Android app. See Devices on the CLI page for how slots are counted and Billing & Plans for how many your plan includes.
• Requests: Vibe and the CLI draw on the same request allowance, and one task usually takes many requests. See Billing & Plans for what counts.

Staying up to date

On Windows, macOS and the Linux AppImage, Vibe checks for a new release and updates itself in place — you do not need to download anything again. The .deb, .rpm and Android builds sit outside that updater; upgrade those by downloading the current file and reinstalling over the top.

Pinning a specific version

Every download link above has a version-pinned twin: replace latest with a release number and the URL is permanent, so it keeps serving that exact build after later releases ship.

https://cdn13.waymore.io/cdn/waymore/vibe/windows/WayMore-Vibe-2.0.0-x86_64.msi
https://cdn13.waymore.io/cdn/waymore/vibe/windows/WayMore-Vibe-2.0.0-aarch64.msi
https://cdn13.waymore.io/cdn/waymore/vibe/macos/WayMore-Vibe-2.0.0-aarch64.dmg
https://cdn13.waymore.io/cdn/waymore/vibe/linux/WayMore-Vibe-2.0.0-x86_64.AppImage
https://cdn13.waymore.io/cdn/waymore/vibe/linux/WayMore-Vibe-2.0.0-aarch64.AppImage

Read latest_version from the release manifest below to find the current release number.

Not every format is published for every release — the .deb and .rpm packages in particular are cut less often than the AppImage. If a pinned URL 404s, that format was not built for that version; try another format or version.

Verifying a download

The Vibe release manifest lists the current version plus the URL, SHA256 and byte size of each installer:

curl -s https://cdn13.waymore.io/cdn/waymore/vibe-manifest.json | jq '.platforms["windows-x86_64"]'
shasum -a 256 WayMore-Vibe-latest-x86_64.msi

The manifest covers the current release and one installer per platform: both Linux AppImages, both macOS DMGs and both Windows MSIs. The .deb, .rpm, the Windows setup.exe and the Android APK are not listed there and have no published checksum, so if you need to verify what you install, use a format the manifest lists.

There is currently no way to verify the Android APK: it has no published checksum, and its debug signature does not prove who built it.

Prefer the terminal?

The WayMore CLI is the same agent for the command line, installed with a single command.

Third-party clients

Subscription plans cover WayMore Vibe, the WayMore CLI and WayMore Chat. Other coding agents pointed at the WayMore API — Claude Code, OpenCode, Cursor and similar — are billed as API usage rather than covered by a subscription. If you need to use one, contact us to enable API billing.

Migration

WayMore Chat provides an OpenAI-compatible API. If you are migrating from OpenAI or Anthropic Claude, the transition requires minimal changes.

From OpenAI

The API is fully compatible with the OpenAI format. You only need to change three things:

Everything else — request/response format, streaming, function calling, message roles — works identically. You can use the OpenAI Python or Node.js SDK by simply overriding the base_url.

From Claude (Anthropic)

WayMore Chat serves the Anthropic Messages API at /v1/messages. Your existing Claude request bodies — including the top-level system field, input_schema tool definitions, and tool_use/tool_result content blocks — are translated automatically, and responses (including streaming events) come back in Anthropic format. The Anthropic SDKs work by overriding the base URL, and x-api-key authentication is accepted. You only need to change:

See the Migrate from OpenAI and Migrate from Claude step-by-step guides for detailed examples and code samples.

Authentication

WayMore Chat supports multiple authentication methods depending on your use case.

API Key Authentication

For programmatic access, include your API key in the Authorization header as a Bearer token. API keys can be scoped with specific permissions and rate limits.

API keys work on the programmatic API surface — chat completions (/v1/chat/completions), model listing (/v1/models), and file upload/download. Account-management endpoints (API keys, billing, usage, content, 2FA) require a signed-in browser session and are not accessible with API keys.

Session Authentication

Browser-based access uses secure HTTP-only cookies managed by NextAuth. Sessions are created automatically when you sign in through the web interface. Session tokens are JWT-based and refresh automatically.

OAuth Providers

Sign in with your existing Google or GitHub account. OAuth accounts are automatically linked if the email address matches an existing account.

Two-Factor Authentication (2FA)

Enhance your account security by enabling TOTP-based two-factor authentication. Once enabled, you will need to provide a 6-digit code from your authenticator app (e.g., Google Authenticator, Authy) each time you sign in. Backup codes are provided during setup for account recovery.

Enable 2FA from Settings → Security in the dashboard: generate the TOTP secret, scan the QR code with your authenticator app, then confirm with a 6-digit code. 2FA setup is only available while signed in to the web interface — it cannot be configured with an API key.

Password Requirements

Passwords must be at least 12 characters and include uppercase letters, lowercase letters, numbers, and special characters. Password changes require your current password for verification.

Chat Interface

The web-based chat interface provides a rich conversational experience with real-time streaming, code highlighting, file attachments, and more.

Conversations

Messages are organized into sessions (conversations). Each session maintains its own context and message history. You can create new sessions, rename them, and switch between them from the sidebar. Sessions are automatically titled based on the first message.

Streaming Responses

Responses stream in real-time using Server-Sent Events (SSE). You see each token as it is generated, providing immediate feedback. Streaming can be used both in the web interface and via the API by setting stream: true.

Code Highlighting

Code blocks in AI responses are automatically syntax-highlighted with a VS Code-inspired theme. Supported languages include JavaScript, TypeScript, Python, Java, Go, Rust, SQL, HTML, CSS, and many more. Each code block includes a copy button and line numbers.

File Attachments

Attach files to your messages for the AI to analyze. Supported file types include images (PNG, JPG, GIF, WebP), PDFs, documents, spreadsheets, and code files. Each file can be up to 50MB, with a maximum of 5 files per message.

Message Ratings

Rate AI responses with thumbs up or thumbs down to help improve response quality. Ratings are tracked and can be used for analytics.

Voice Input

Use the microphone button to dictate messages using speech-to-text. Voice input uses a WebSocket connection for real-time transcription.

API Keys

API keys provide programmatic access to the WayMore Chat API. Each key can be configured with specific permissions, rate limits, and IP restrictions.

Creating Keys

Create API keys from the web dashboard while signed in (key management endpoints use your browser session — an API key cannot create other keys). The number of keys you can create depends on your subscription plan: 2 for Free, 5 for Starter, 20 for Pro, 100 for Enterprise. The full key value is shown only once at creation — store it securely.

Permissions

API keys support granular permissions to restrict what operations the key can perform:

Rate Limits

Each key can have configurable rate limits: requests per minute (RPM), daily token limit, and monthly token limit. When a limit is exceeded, the API returns a 429 Too Many Requests response.

IP Whitelisting

Restrict API key usage to specific IP addresses or CIDR ranges. Requests from non-whitelisted IPs will be rejected with a 403 Forbidden response.

Key Lifecycle

API keys can be in one of three states: Active, Revoked, or Expired. You can regenerate a key to get a new secret while keeping the same configuration. Revoked keys can be reactivated if needed. Keys can also be set to auto-expire after a specified number of days.

Models

WayMore Chat provides access to AI models for text generation, analysis, and conversation.

Available Models

Query the models endpoint to see every model available to your account and plan. It accepts an API key or a signed-in browser session.

• Astrape — WayMore Astrape — fast 1M-context model with always-on reasoning. Open to all plans.
• Valkyrie 5 — WayMore Valkyrie 5 — flagship 1M-context model with always-on reasoning. Open to all plans.
• Waymore-A1-Extract — WayMore A1 Extract (POC) — structured extraction: a forced single tool is answered through strict JSON Schema mode server-side.
• Waymore-A1-Instruct-1011 — WayMore A1 Instruct — 1M context, 64K max output.
• Waymore-A1-Instruct-1012 — WayMore A1 Instruct 1012 — 1M context, 128K max output.
• Waymore-A1-Turbo — WayMore A1 Turbo — fast chat, non-thinking by default; honors an explicit thinking flag. 1M context.
• Zeus — WayMore Zeus — preview task routing with focused execution profiles for coding, interface design, research and review.
• Atlas 5 (Restricted) — WayMore Atlas 5 (Restricted) — most-capable premium model: 2M context with extra-high reasoning for complex, multi-step agentic tasks. Opt-in only — apply for access. (access is opt-in — contact support to be enabled)

Pass any non-restricted model ID as the model field on a chat completion request. If you have no preference, Waymore-A1-Instruct-1011 is the general-purpose default, and WayMore auto-routes each request to the best backend for the prompt.

Checking an API key

To test whether a key still works, make a real request and read the status code: 200 means the key is valid, 401 means it is wrong, revoked or expired, 403 means it is valid but not permitted for that model, and 429 means it is valid but rate-limited. Keys issued in the dashboard begin with sk-.

Do not test a key against /api/chat/models. That endpoint is internal to the web interface and authenticates with a browser session only, so it returns 401 for every API key — valid or not.

Model Parameters

When sending a chat completion request, you can configure the following parameters:

Function Calling

Extend the model's capabilities by defining custom functions (tools) that it can invoke during a conversation. The API supports both OpenAI and Anthropic tool formats — use the OpenAI format on /v1/chat/completions and the Anthropic format on /v1/messages.

How It Works

You define tools in your API request. When the model determines a tool should be used, it returns the function name and structured arguments. You execute the function on your side, then send the result back in a follow-up request. The model then generates a final response using the tool output.

Supported Formats

Each endpoint responds in its own format:

Tool Definitions

Controlling Tool Use

Parallel Tool Calls

The model can call multiple tools in a single response. For example, asking "What's the weather in Athens and London?" may produce two tool calls in the same response. Each has a unique ID — submit results for all of them before making the next completion request.

Multi-Turn Flow

The complete function calling flow involves three steps:

1. Send a request with tools defined. The model returns tool calls with function names and arguments.
2. Execute the function(s) on your side using the arguments provided by the model.
3. Send the results back (as tool role messages in OpenAI format, or tool_result content blocks in Anthropic format). The model generates a natural language response.

See the API Reference for complete request/response examples, and the Function Calling Guide for a step-by-step tutorial.

Usage & Analytics

Monitor your API consumption, track costs, and analyze usage patterns through the usage dashboard or API endpoints.

Usage Dashboard

The web dashboard provides visual charts showing your request volume, token consumption, cost breakdown, and error rates over time. Filter by date range, API key, model, and request status.

Querying Usage via API

Use the usage API to programmatically retrieve your consumption data. Supports filtering by time period, API key, and model.

Metrics Tracked

Alerts

Set up usage alerts to receive email notifications when you approach or exceed token or cost thresholds.

Per-Key Usage

View usage statistics for individual API keys, including daily breakdowns and request history:

Billing & Plans

Every plan covers the WayMore CLI, WayMore Vibe and WayMore Chat (the web app). Pick a tier by how many requests you make and how many machines you work on.

Individual plans

Prices are per month. Devices is how many sign-ins can be active at once (see WayMore CLI for how devices are paired and counted).

Team plans

Team plans are billed per seat. One organization holds one subscription and one payment method, and each member gets the allowances below individually. Team Business and Team Enterprise also add a shared pool for members who reach a limit. See Teams for how the pool works and how to set up a team.

How the request limits work

The monthly figure is your allowance. The daily and weekly figures limit how fast you can use it, and all three apply at once. On paid plans the weekly figure equals the monthly one, so it only matters in a week that spans two calendar months.

Pro, for example, includes 1,250 requests a month. You could use 938 in one day and the remaining 312 the next, and then have none left until the next month.

The limits reset at fixed times in UTC, not on a rolling basis and not on your billing date:

Reaching a limit returns 402, and requests are refused until it resets. See Errors & limits for the response and what to do about it.

What counts as a request

One request is one call to the model, not one prompt or one task. An agent usually makes many calls to finish a single task:

• Each time the agent uses a tool, such as reading a file or running a command, it sends the result back in a new request.
• Each subagent makes its own requests.
• Background calls the agent makes for itself, such as compacting a long conversation, count too.

Over the 30 days to November 2026, a typical customer session used about 36 requests, and one session in ten used more than 120. At that rate, Pro's 1,250 requests a month cover about 35 typical sessions. Long tasks and heavy use of subagents use up the allowance faster.

Failed requests do not count. The WayMore CLI and WayMore Vibe share one allowance. WayMore Chat is counted separately: Free includes 25 chat requests a day and 300 a month, and paid plans have no chat limit.

Starting a paid plan

Paid plans are activated through checkout in the billing dashboard, where you enter card details and the subscription is created before the plan is applied to your account.

Changing plans from the API is deliberately limited: you can always move down a tier or between free plans, but a self-service upgrade to a paid plan is refused. Upgrades go through checkout, or through sales if you pay by bank transfer.

Payment methods

Payments are processed by Braintree. Card details are entered into Braintree's own hosted form and never reach WayMore's servers — we store only the resulting token. You can keep more than one instrument on file and set a default for recurring charges.

Invoices

Invoices for past payments are listed in the billing dashboard, itemised by plan charge, seat count where it applies, and any prorated adjustment from a plan change. Billing pages use your signed-in browser session; there is no API-key access to billing data.

Cancellation

Cancel at any time from the billing dashboard. Cancellation takes effect at the end of the current billing period and you keep your plan's features until then. Cancelling a team subscription ends it for every seat in the organization.

Teams

An organization puts a group of people on one subscription and one invoice, while each member keeps their own devices and their own request allowance.

How a team plan works

An organization holds one subscription — a Team plan from Billing & Plans — and pays as one customer. Every member is a seat, and the price is per seat.

Allowances are per member. On Team Business each member gets 2,500 requests a month and up to 20 devices of their own, and one heavy user cannot consume a quieter colleague's allowance.

Shared pool

Team Business and Team Enterprise also include a shared pool. When a member reaches one of their own limits, their requests draw on the pool instead of being refused. Team Lite, Team Standard and Team Starter have no pool.

Each month's pool is 25% of the team's combined monthly allowance, so 10 Team Business seats get 6,250 requests (10 × 2,500 × 25%). Unused pool carries over to the next month, up to a total of twice the monthly pool. A team with no requests in the previous month gets no carry-over.

When a member's allowance and the pool are both used up, their requests are refused with 402. The team console shows the pool balance, and Boost Packs bought there add more requests to it.

Creating an organization

From the billing dashboard, give the organization a name, a short URL slug and a billing email. You become its owner. A new organization has no subscription yet — subscribe to a Team plan to start adding members.

One person can own at most 5 organizations. Being a member of others does not count towards that.

Roles

There is exactly one owner at a time. To hand an organization over, transfer ownership — an owner cannot simply be removed, and an account that still owns an organization cannot be deleted.

Inviting members

Invite people by email address. They receive a link and, after signing in, join with the role you picked. Invite links expire after 7 days by default (30 at the most), and you can revoke one before it is used.

An invite is refused when:

• the link has already been used, or was revoked
• it has expired
• the signed-in email does not match the address the invite was sent to
• the organization has no free seat

That last one is the common case. Raise the seat count first, then re-send.

Seats

The seat count is what you are billed for, and it caps membership: adding a member when every seat is taken is refused. Seats can be increased at any time. They cannot be reduced below the number of members you currently have — remove members first.

Working in more than one organization

If you belong to several organizations that each pay for a Team plan, the billing page shows which one you are currently using and lets you switch. Your CLI requests are then billed and metered against that organization.

If you have a personal plan as well, you can switch back to it the same way. Members of a single organization see no picker — there is nothing to choose between.

Who to ask about billing

Only the owner can see invoices and change the plan. Members see which organization they belong to, and the owner's email address for billing questions.

Content Library

Save and organize content from your AI conversations into a personal library. Content items include images, videos, files, code snippets, conversations, and notes.

Content Types

Organizing Content

Use tags, favorites, and collections to keep your library organized. Content items support full-text search across titles, descriptions, and tags. Filter by type, source, favorite status, and archive status.

Storage

Each account has a storage limit based on your subscription plan. Track your storage usage through the content statistics endpoint. File uploads are limited to 50MB per file.

Collections

Group related content items into collections for better organization. Each collection can have a custom name, description, color, and icon.

Creating Collections

Create collections from the content library page. A default collection is created automatically for each new account. Collection endpoints authenticate with your signed-in browser session — they are not accessible with API keys.

Managing Items

Add or remove content items from collections in the content library. An item can belong to multiple collections. Default collections cannot be deleted.

Canvas Documents

Canvas is a collaborative document editor linked to chat sessions. Use it to draft, edit, and iterate on content alongside your AI conversations.

How Canvas Works

Each chat session can have an associated canvas document. The canvas stores content alongside the conversation, allowing you to work on a document while discussing it with the AI. Canvas documents support version history with up to 50 snapshots.

Live Collaboration

Multiple collaborators can edit the same canvas in real time. Presence indicators show who is currently editing, along with their cursor position. Changes are synced instantly through our WebSocket infrastructure.

Version History

Canvas automatically tracks changes through version snapshots. You can view, compare, and restore previous versions. When the maximum of 50 versions is reached, the oldest versions are automatically removed.

Canvas endpoints authenticate with your signed-in browser session — they are not accessible with API keys.

Files & Uploads

Upload files for AI processing or attach them to chat messages. The platform supports a wide range of file types.

Supported File Types

Upload Limits

Maximum file size is 50MB per file. Chat attachments are limited to 5 files per message. Files are processed through the LLM backend for AI analysis.

Uploading directly to the content library (POST /api/stuff/upload) is a dashboard feature — it authenticates with your signed-in browser session and is not accessible with API keys.

Security

WayMore Chat is built with security as a priority. Here is an overview of the security measures in place.

Transport Security

All traffic is encrypted with TLS (HTTPS). HTTP Strict Transport Security (HSTS) is enforced with a max-age of 2 years, includeSubDomains, and preload. API keys and session tokens are only transmitted over encrypted connections.

Security Headers

The platform sets comprehensive security headers on all responses:

Account Security

Protect your account with strong passwords (minimum 12 characters), two-factor authentication, and API key IP whitelisting. Session tokens use secure, HTTP-only cookies that are not accessible to JavaScript.

Rate Limiting

Authentication endpoints are rate-limited to prevent brute-force attacks. API keys support configurable per-minute, daily, and monthly rate limits. Excessive requests return a 429 status code with a Retry-After header.

Errors & limits

Every failure comes back as JSON with an error field. The status code tells you whether to retry, fix the request, or change something on your account.

Status codes

403 in detail

Two different 403s come from device control, and they are not interchangeable:

• device_limit_reached — you already have as many active sign-ins as your plan allows. Revoke a key you no longer use, or move up a tier.
• device_blocked — an administrator blocked this specific device. Signing in again will not clear it; the block survives re-pairing by design. The response carries a block_id you can quote to your administrator.

Device counts per plan are in Billing & Plans.

402: a plan limit was reached

402 means you have used up one of your plan's request limits. Retrying will not help until that limit resets: daily limits reset at 00:00 UTC, weekly limits on Monday and monthly limits on the 1st. To keep working sooner, upgrade. See Billing & Plans for what counts as a request.

The response names the limit and says how long until it resets, in retry_after_seconds and in a Retry-After header. actions lists the ways to get more, such as an upgrade.

On Team Business and Team Enterprise, a member gets this only after the shared pool is used up as well.

A 402 with the code org_budget_exceeded is different: your organization has reached the monthly budget its admin set. An owner or admin can raise or remove the budget in the team console.

Current consumption for each window is shown under Usage.

429: too many requests at once

429 means one API key sent too many requests in a short time. It is separate from your plan's limits and clears by itself, usually within a minute. The Retry-After header says how many seconds to wait.

For automated retries, use exponential backoff rather than a fixed delay: wait a second, then two, then four, up to about a minute, and add a little random jitter so a fleet of workers does not retry in lockstep. Retry only on 429, because a 402 will keep failing until the limit resets.

Checking whether a key is the problem

Make a real request and read the status: 200 means the key works, 401 means it is wrong or revoked, 402 means it is valid but a plan limit has been reached, 403 means it is valid but not permitted for what you asked, and 429 means it is valid and merely throttled.

Do not test a key against /api/chat/models. That endpoint belongs to the web interface and accepts only a browser session, so it answers 401 for every API key — working or not.