the1812/warp-offline

Warp Offline Fork

This repository is a local offline-oriented Warp fork. The original upstream README has been kept as README.upstream.md.

Offline Mode

The warp-oss binary starts in local-only services mode. In this mode the app uses ServicesMode::LocalOnly, local-only server/Oz configuration, and no telemetry, crash reporting, autoupdate, or statically bundled MCP OAuth configuration.

Offline mode is intended to keep Warp useful as a local terminal while removing Warp-authored online service calls. User-run shell commands are outside this boundary: commands such as curl, ssh, git fetch, or package managers still use the operating system network stack if the user runs them in the terminal.

When offline mode is active, these product surfaces are hidden, disabled, or guarded by local-only runtime checks:

• Warp account, sign-up, upgrade, billing, referrals, Team, Workspace, and Cloud Platform API key flows.
• Warp-hosted Agent, Cloud Agent, Oz, cloud conversation, orchestration, managed secret, and Warp-hosted AI entry points.
• Warp Drive, cloud objects, cloud object sync/listener/update queues, Drive search, Drive panel actions, cloud workflows/notebooks/env vars, and shared cloud object operations.
• Session sharing, RTC/WebSocket subscriptions, Firebase/OAuth-backed login flows, MCP OAuth/static cloud configuration, and remote cloud metadata polling.
• Telemetry, active-usage flushing, Sentry/minidump crash reporting, automatic update polling, online changelog fetches, and remote fallback asset/font fetching.
• Privacy/secret-redaction online surfaces that depend on cloud policy or online service behavior.

The network policy rejects Warp-managed public HTTP/HTTPS/WS/WSS requests in local-only mode and allows loopback URLs such as localhost, 127.0.0.1, and ::1. Bundled and file-like assets are still allowed. Manual SSH and remote-server workflows are preserved when they use an already available local binary or user-run shell commands; automatic Warp-managed remote-server downloads are disabled.

Local AI

Local AI is the supported AI path for this fork. It is configured from the top-level Settings page named Local AI, not from the Warp Agent settings page. The settings are stored under:

[agents.local_openai]
enabled = true
base_url = "http://127.0.0.1:1234/v1"
api_key = ""
command_model = "google/gemma-4-e4b"
prediction_model = "google/gemma-4-e4b"
timeout_ms = 8000

The provider uses an OpenAI-compatible /v1/chat/completions API. The API key is optional; when present it is sent as Authorization: Bearer ... to the configured endpoint.

Local AI currently backs local command-oriented features:

• AI command search and natural-language command generation.
• Workflow metadata autofill.
• AM query prediction and next-command LLM fallback.

It does not use Warp credits, Warp account state, workspace policy, Warp-hosted BYOK, Cloud Agent, Oz, or the hosted Warp Agent service. In local-only mode, command-search entry points are gated by agents.local_openai.enabled, while the request path still validates the base URL and model settings before calling the provider.

Platform Status

This offline fork has only been tested on Windows 10. Windows 11, macOS and Linux have not been validated, and platform-specific packaging, update, remote-server, and GUI paths may still need follow-up work before those platforms can be considered supported.