You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
--ollama-url <URL> Ollama server URL (default: http://localhost:11434)
106
-
--max-turns <N> Max turns per conversation (default: 100)
70
+
# windows
71
+
.\target\release\clawcr onboard
107
72
```
108
73
109
-
### Supported Providers
110
-
111
-
| Provider | Backend | How to activate |
112
-
|----------|---------|-----------------|
113
-
|`ollama`| Ollama (local) |`--provider ollama` or auto when no `ANTHROPIC_API_KEY`|
114
-
|`anthropic`| Anthropic API | Set `ANTHROPIC_API_KEY` env var |
115
-
|`openai`| Any OpenAI-compatible API |`--provider openai` + `OPENAI_BASE_URL`|
116
-
|`stub`| No real model (for testing) |`--provider stub`|
117
-
118
-
## 🤔 Why Rebuild in Rust
119
-
120
-
Claude Code has excellent engineering quality, but it's a **complete product**, not a reusable runtime library. UI, runtime, tool systems, and state management are deeply intertwined. Reading the source teaches a lot, but extracting parts for reuse is nontrivial.
121
-
122
-
This project aims to:
123
-
124
-
-**Decompose** tightly coupled logic into single-responsibility crates
125
-
-**Replace** runtime constraints with trait and enum boundaries
126
-
-**Transform** "only works inside this project" implementations into **reusable agent components**
127
-
128
-
## 🎯 Design Goals
129
-
130
-
1.**Runtime first, product later.** Prioritize solid foundations for Agent loop, Tool, Task, and Permission.
131
-
2.**Each crate should be self-explanatory.** Names reveal responsibility, interfaces reveal boundaries.
132
-
3.**Make replacement natural.** Tools, model providers, permission policies, and compaction strategies should all be swappable.
133
-
4.**Learn from Claude Code's experience** without replicating its UI or internal features.
0 commit comments