cargo bundle package + logo + readme WIP

Cargo.toml

@@ -2,6 +2,33 @@
 name = "weave"
 version = "0.1.0"
 edition = "2021"
+description = "A tool for collaborative generative writing."
+
+[package.metadata.bundle]
+name = "Weave"
+identifier = "dev.mdegans.weave"
+resources = ["resources"]
+copyright = "2024, Michael de Gans"
+category = "public.app-category.productivity"
+long_description = """
+A tool for collaborative generative writing. It supports multiple generative
+backends such as OpenAI and LLaMA. Stories can branch, allowing for multiple
+paths through a story. It is designed to be used by writers, game developers,
+or anyone who needs to generate text.
+"""
+icon = [
+    # "resources/icon.32.png",
+    # "resources/icon.32@2x.png",
+    # "resources/icon.64.png",
+    # "resources/icon.64@2x.png",
+    # "resources/icon.128.png",
+    # "resources/icon.128@2x.png",
+    "resources/icon.256.png",
+    # "resources/icon.256@2x.png",
+    "resources/icon.512.png",
+    "resources/icon.512@2x.png",
+    # "resources/icon.1024.png",
+]
 
 [dependencies]
 serde = { version = "1.0", features = ["derive"] }
@@ -40,6 +67,7 @@ openai = [
     "dep:futures",
     "dep:keyring",
     "dep:tokio",
+    "tokio/rt-multi-thread",
 ]
 ollama = ["generate", "dep:ollama-rs"]
 # TODO: Claude does not yet have a good rust library. Will have to use reqwests

README.md

@@ -1,16 +1,34 @@
 # `weave` multiversal writing tool
 
+![Weave Icon](/resources/icon.inkscape.svg)
+
 Weave is a "multiversal" generative tree writing tool akin to [`loom`](https://github.com/socketteer/loom). It supports multiple generative backends such as:
 
-- [x] [`drama_llama`](https://github.com/mdegans/drama_llama) - llama.cpp wrapper supporting all llama.cpp models
-- [ ] OpenAI models
-  - [ ] GPT 3.5 completions
-  - [ ] Shim for GPT 4+ chat completions API.
-- [ ] Anthropic models
+- ✅ [`drama_llama`](https://github.com/mdegans/drama_llama) - llama.cpp wrapper supporting all llama.cpp models
+- ✅ OpenAI models
+  - ✅ Shim for GPT 3.5+ chat completions API, including GPT 4o.
+- 🔲 Anthropic models
+
+## Features
+
+Notable features:
+
+- **Live switching of backends** - It's possible to generate part of a story
+  with OpenAI and another part with LLaMA -- all without restarting the app.
+- **Streaming responses** - It's possible to cancel generations in progress --
+  both local and online.
+- **Live editing** - It's possible to edit posts during generation, but not to
+  add or remove nodes, so you need not wait for generation to complete to tweak
+  the text to your liking. New tokens are always appended to the end.
 
-# Features
+Coming soon:
 
-The goal of `weave` is feature parity with [`loom`](https://github.com/socketteer/loom?tab=readme-ov-file#features).
+- Fine-grained support over sampling for local models and potentially remote as
+  well for backends returning logprobs. The backend code is already written in
+  `drama_llama` but this is not exposed.
+- Keyboard shortcuts.
+
+Additionally, one goal of `weave` is feature parity with [`loom`](https://github.com/socketteer/loom?tab=readme-ov-file#features).
 
 - ☑️ Read mode
   - ✅ Linear story view
@@ -28,9 +46,33 @@ The goal of `weave` is feature parity with [`loom`](https://github.com/sockettee
   - 🔲 'Visited' state
 - ☑️ Generation
   - 🔲 Generate N children with various models (currently one a time).
-  - ✅ Modify generation settings
+  - ☑️ Modify generation settings (Complete for OpenAI but not yet from LLaMA)
 - ☑️ File I/O
   - ✅ Serializable application state, including stories, to JSON.
-  - 🔲 Open/save trees as JSON files
+  - ✅ Open/save trees as JSON files
   - 🔲 Work with trees in multiple tabs
   - 🔲 Combine multiple trees
+
+# Notable issues
+
+- On some platforms (like MacOS) the Weave icon will change to an `e` shortly
+  after launch. See [this
+  issue](https://github.com/emilk/egui/issues/3823#issuecomment-1892423108) for
+  details.
+- With each new generation, all tokens need to be injested again with most
+  backends. This is solvable with `drama_llama` (longest prefix cache) but not
+  for the OpenAI API. So for OpenAI, it's recommended to generate larger posts.
+  The system prompt is customizable so you can tweak the agent's instructions on
+  verbosity.
+- It is not currently possible to have a scrollable viewport so it's
+  recommended to collapse nodes if things get cluttered. This is because the
+  nodes are implemented with [`egui::containers::Window`](https://docs.rs/egui/latest/egui/containers/struct.Window.html) which ignore scrollable areas. This is fixable
+  but not easily and not cleanly. When it is resolved the central panel will be
+  split into story and node views.
+- For the same reason as the above, nodes may obscure the generated text view.
+- The `drama_llama` backend will crash if the model's output is not valid
+  unicode. This will be fixed. If this happens, go to settings, switch backends,
+  and then switch back `drama_llama`.
+- The BOS token is not added for the `drama_llama` backend. This will be added
+  as an option and enabled by default since most models expect it. Generation
+  will still work but the quality may be affected.

resources/COPYRIGHT.md

@@ -0,0 +1,7 @@
+# Icon Copyright
+
+The app Icon was generated by Bing's Copilot and Dall-E 3. It was then post
+processed in inkscape to create a vector version then rasterized at several
+different resolutions.
+
+Because the icon is generated, it effectively falls into the public domain.

resources/icon.128@2x.png

@@ -0,0 +1 @@
+icon.256.png

resources/icon.256@2x.png

@@ -0,0 +1 @@
+icon.512.png

resources/icon.32@2x.png

@@ -0,0 +1 @@
+icon.64.png

resources/icon.512@2x.png

@@ -0,0 +1 @@
+icon.1024.png

resources/icon.64@2x.png

@@ -0,0 +1 @@
+icon.128.png