Walkthroughs for installation, integration, billing, and the common errors people hit on their first few sessions. Each topic has a deep-link anchor — share the URL with a teammate and they'll land on the right answer.
The Terminal app is a small, code-signed application that runs on your Windows, Mac, or Linux machine. It lets AI Expedite's central agents dispatch coding tasks (Claude Code, Codex, builds, tests) to your computer using your own subscription and toolchain.
The first time the terminal app runs a command outside the default allowlist, your OS will show a native approval dialog with the exact command. You can approve once, approve as a pattern, or deny. The agent cannot run anything you have not explicitly approved.
If you don't see the connect button after install, see the Troubleshooting the Terminal topic below.
GitHub is the substrate AI Expedite builds on — every feature, branch, PR, and code analysis pass routes through it. You need to connect a GitHub account before agents can run coding work.
AI Expedite requests repository-scoped access — never org-wide read. The GitHub App permissions are visible during the install flow and listed on our Security page.
AI Expedite can route Claude Code work through your existing Claude Pro, Max, or Team subscription. This is the cheapest way to run agentic coding at any meaningful volume.
If the worker doesn't show Claude Code as ready, check that the CLI is on the system PATH the terminal app is using. We never see or proxy your Anthropic credentials — the CLI handles auth directly with Anthropic.
AI Expedite supports both the Codex CLI (subscription-routed) and the OpenAI Batch API (metered) for async coding work. The orchestrator routes to the cheaper viable option per job.
Model Context Protocol (MCP) servers let your agents reach external tools and data sources — internal APIs, databases, custom search indexes, anything that exposes an MCP endpoint.
MCP server calls flow through AI Expedite's request layer, so they get the same logging, rate limiting, and audit trail as built-in tool calls.
Workspaces have a small set of roles, enforced server-side via Firestore Security Rules. Membership is managed in Settings → Workspace → Members.
Code analysis and scheduled activities are billed to the workspace owner, not the individual member running the work. That keeps cost predictable for teams while letting each engineer drive agents from their own account.
If you're trying to link your GitHub account but you're receiving the error 'Email already in use', that typically indicates that you have signed into AI Expedite with multiple accounts.
This means the AI Expedite GitHub App is installed on a different account than the one that owns the repo, or the repo is not in the app's configured allowlist. Open GitHub → Settings → Applications → Installed GitHub Apps → AI Expedite, and confirm the repo is listed. If not, click Configure and add it.
Long-lived GitHub installation tokens are refreshed automatically. If you see this error persistently, disconnect the GitHub integration from Settings → Integrations and reconnect. The reconnect refreshes the underlying installation token.
The action you tried to perform requires a permission the AI Expedite GitHub App does not currently hold. This is most common when a new feature requires a scope that existing installations don't yet grant. Open the GitHub App settings → Permissions and approve the new permissions when prompted.
AI Expedite uses credits to meter AI execution. Browsing the app, editing documents manually, and managing billing settings do not consume credits.
If you hit your included monthly credit limit, AI Expedite pauses new AI work by default. Auto-Reload lets you set a monthly spending cap and automatically tops up credits when you run out. Find it in Settings → Billing → Auto-Reload. The cap is hard — once you hit it, work pauses again until the next month or until you raise the cap.
Prepaid top-ups are a one-time purchase that adds credits to your account immediately. Top-up credits roll over for 60 days, then expire. Included monthly credits do not roll over.
Agents fail for predictable reasons. The session always surfaces in the workspace with the full transcript, the build / test output, and the current state — failed runs are never silently retried.
Code Analysis builds a structured index of a repo — file structure, exports, call graph, type information, conventions in use. Agents reference the index to reason about code in context: when Claude Code edits a function, the orchestrator knows about every call site; when the reviewer agent reads a diff, it knows what tests cover the changed lines.
Code Analysis is a workspace-owner cost: it bills to the workspace, not the individual member who enabled it. The cost is roughly proportional to repo size and commit volume. For very large monorepos, consider scoping analysis to the directories your team actively works in.
Ship still works without Code Analysis — agents fall back to live reads against the repo. The downside is slower runs and less informed review. Discover features that depend on effort estimation, and Launch features that depend on codebase context, both benefit substantially from having analysis enabled.
Security review is in limited preview: it is open to selected workspaces only, and its card in Settings → Discover shows it when your workspace can turn it on.
Security review is a nightly Discover job that reviews your workspace's own code for security flaws. Each night it reads what changed on your repositories' default branches, plus a rotating set of areas reviewed in full. It re-checks the findings it reported before and files a report in Library → Reports. A log of what it has reviewed is kept in the workspace, so each night picks up where the last one stopped.
The review runs on the coding agent that Ship routing picks for Discover, such as Claude Code or Codex, on a computer that belongs to the person whose account pays for the run. It uses that computer's own sign-in.
The agent runs read-only, enforced by the computer, not merely asked to behave. Claude Code gets only its Read, Grep and Glob tools, with no shell, no web access and no MCP servers. Codex runs in its read-only sandbox, where writes and network access are blocked by the operating system. A coding agent with no such mode is skipped, and so is Codex on Windows. If no routed agent can run read-only, the report says so and nothing is reviewed that night.
The reserved computer needs an up-to-date Terminal app; an older one is asked to update. The job never writes an exploit and never sends anything to a deployed service. It describes each flaw in prose, along with the fix and a test to add.
Nothing is ever installed silently. The job card in Settings → Discover lists the optional tools on each computer. The computer's owner can choose Install…, which runs the same setup the computer checklist uses, limited to those tools, and only after the owner approves it. Everyone else is shown who to ask. Without the tools, the night skips only the part that needs them.
Reading security alerts needs the GitHub App's Dependabot alerts, code scanning alerts and secret scanning alerts read permissions. Until an owner of your GitHub account approves them, those reads are skipped, and the job card shows Approve the new GitHub permissions with a link to approve.
The review runs under each vendor's default safeguards. If a model refuses the request, that night's review counts as not done rather than clean. The next agent in the ranking tries instead, and the next night reviews the same changes again.
If your team does security work that the default safeguards get in the way of, you can apply for Anthropic's Cyber Verification Program or OpenAI's Trusted Access for Cyber for your own organization. Approval applies to the accounts your coding agents sign in with.
Confirm you're signed in to the same AI Expedite account in the app and in the browser. If the Connect to cloud button is greyed out, sign out of the app, sign back in, and retry. Corporate networks sometimes block the outbound connections — check that `*.aiexpedite.com` is reachable.
Open the app's Status tab. Each capability (Claude Code, Codex, build, test) shows a green / red indicator. Red means the underlying tool isn't on the PATH the app uses, or isn't authenticated. Install or authenticate the tool, then click Re-detect.
If a command should be prompting for OS approval but isn't, it may already be on the allowlist (intended) or your OS may be configured to suppress dialogs (not intended). Check the Allowlist tab in app settings — anything you've previously approved as a pattern runs without re-prompting.
The app enforces a per-command timeout (default 10 minutes). Long-running jobs (large builds, full test suites) can exceed it. Raise the timeout in app settings for the affected repo, or split the command into smaller steps.