Skip to content
black12-agPublic

About

One terminal, all your CLI subscriptions — Claude Code, Codex & Gemini in one chat. No login, no API keys. Fork of Warp OSS.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

2,162 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Weft — one terminal, all your CLI subscriptions

Latest release Downloads Stars License: MIT/AGPL

In weaving, the weft is the thread that crosses the warp. A fork of Warp's open-source terminal that weaves Claude Code, OpenAI Codex, and Google Gemini into one chat.

Its AI chat runs on your own local CLI agents using the subscriptions you already pay for.

No Warp account. No sign-in. No API keys needed. No Warp servers. Ever.

⭐ If Weft is useful to you, star the repo — it helps others find it. · ⬇️ Download the latest release

Weft is CLI-only today — models run through your CLI subscriptions, so you don't need to pay for API keys. (Bringing your own API key as an alternative to a CLI subscription is planned but not in this version.)

What you get

  • One chat, every model — pick any Claude / Codex (GPT) / Gemini model from the model button, switch mid-conversation, and the new model continues exactly where the last one stopped. Cross-provider memory is built in.
  • Thinking levels — every model exposes its reasoning range (low → max / minimal → xhigh). Pick the speed↔depth trade-off per model from the composer's model popup.
  • Live streaming — answers and the model's thinking type out live; the thinking auto-collapses into a "Thought for Xs" dropdown when the answer lands. Token usage is shown after each reply.
  • Agentic Codex — Codex can read files and run commands in your current project (read-only sandbox) and shows each step live.
  • Fast — Claude ~4s, Codex ~6s for a short reply. Fast mode skips each CLI's heavy MCP/plugin/skill loading; type tools on in the chat to load your full personal MCP + skills config when you need it (tools off to go back).
  • A brain — every exchange is saved as markdown (to your Obsidian vault's Weft/ folder if you have one, else ~/.weft/brain/), one note per project, and recalled automatically in later sessions. All models share it. memory on / memory off / memory clear.
  • Connect a CLI from Settings — Settings → Agents → Third party CLI agents shows each CLI (installed? signed in?) with a Connect button that runs the provider's own login (claude auth login, codex login, gemini). Credentials go straight to the provider; this app never sees them.

Install

macOS — Intel & Apple Silicon (one Universal build)

One command (recommended — installs with no Gatekeeper warning):

curl -fsSL https://raw.githubusercontent.com/black12-ag/weft/master/install-mac.sh | bash

Or download Weft.dmg from the latest release, drag Weft to Applications, then right-click → Open (ad-hoc signed, so macOS asks once) — or run xattr -dr com.apple.quarantine /Applications/Weft.app.

Linux — build from source (one command)

curl -fsSL https://raw.githubusercontent.com/black12-ag/weft/master/install-linux.sh | bash

Builds on Debian/Ubuntu, Fedora, and Arch. Prebuilt .deb / .rpm / AppImage coming soon.

Windows — build from source (one command, PowerShell)

irm https://raw.githubusercontent.com/black12-ag/weft/master/install-windows.ps1 | iex

Prebuilt .exe installer coming soon.

Then

Install / sign in to at least one CLI (each uses its own subscription):

  • Claude Code: npm i -g @anthropic-ai/claude-code then claude auth login
  • Codex: npm i -g @openai/codex then codex login
  • Gemini: npm i -g @google/gemini-cli then run gemini once

Open Weft → use the chat → pick a model from the model button → go. Manage your MCP servers and see each model's specs (intelligence / speed / cost + $ per 1M tokens) in Settings → Third party CLI agents.

Build from source

Requirements: macOS, Xcode (with the Metal toolchain), Rust (rustup — the repo pins the toolchain), protoc.

git clone <this repo>
cd warp
./install.sh

The script builds the release app, packages Weft.dmg, and installs to /Applications. See install.sh for the individual steps.

Customize it

It's all source — the local-CLI integration lives mainly in:

  • app/src/ai/agent/api/impl.rs — runs the CLIs, streams thinking/answers, transcript memory, the Obsidian brain, tools/memory chat commands.
  • app/src/ai/llms.rs — which models each CLI exposes (add your own here).
  • app/src/settings_view/cli_agents_page.rs — the Connected CLIs settings section.

Add another CLI by extending build_local_cli_invocation (how to call it) and local_cli_llm_choices (which models to list).

Credits — what's Warp, what's Weft

Weft stands on Warp's work, and credits it fully. The entire terminal is Warp's open source: the terminal engine, the whole UI and UX, the blocks/rendering, the agent chat surface, the settings system, the model-picker components, and the client/agent backend. None of that is ours — Warp built it, and it's excellent.

What Weft changes is one layer: where the AI comes from. Warp's AI chat talks to Warp's cloud (account + credits). Weft rewires that layer so the same chat runs on your own local CLI agents — the subscriptions you already pay for — with no account, no login, no API keys, and no calls to any server. Concretely, Weft adds/modifies:

  • Local-CLI execution — the agent chat runs claude / codex / gemini locally instead of Warp's servers (app/src/ai/agent/api/impl.rs).
  • All models + thinking levels, no server — every model each CLI offers, with per-model reasoning/effort (app/src/ai/llms.rs).
  • Live streaming of the answer and the model's thinking, with an auto-collapsing "thought" block and token counts.
  • One shared conversation across providers — switch Claude↔Codex↔Gemini mid-chat and it continues where you left off.
  • A persistent brain — every exchange saved as markdown (Obsidian vault or ~/.warposs/brain/) and recalled across sessions.
  • Fast mode — skips each CLI's heavy MCP/plugin/skill scan for terminal-speed replies (tools on loads them when needed).
  • Connect-a-CLI settings — Settings → Third party CLI agents shows install/login status and a Connect/Disconnect button per CLI.
  • No-login build — sign-in prompts and credit banners removed (the skip_login feature and related gates).

This project is actively updated — expect frequent changes as the CLI integration, memory, and settings evolve.

Links & credits

Author — Munir Ayub

Full-stack, mobile & AI developer.

License

Weft is a fork of Warp's open-source terminal and keeps Warp's licensing in full — see LICENSE-MIT and LICENSE-AGPL. All original terminal/UI/UX/backend code is © Warp (Denver Technologies, Inc.). Weft's modifications (the local-CLI AI layer described above) are released under the same terms. If you distribute Weft or a build of it, you must keep these licenses and this attribution.

About

One terminal, all your CLI subscriptions — Claude Code, Codex & Gemini in one chat. No login, no API keys. Fork of Warp OSS.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages