Skip to content

Commit f5a51e8

Browse files
Vadim MitroshkinVadim Mitroshkin
authored andcommitted
README update
1 parent 536a2fe commit f5a51e8

1 file changed

Lines changed: 40 additions & 15 deletions

File tree

README.md

Lines changed: 40 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -89,25 +89,50 @@ export SUPERCODER_API_KEY="sk-or-..."
8989
export SUPERCODER_MODEL="openai/gpt-4o"
9090
```
9191

92-
**Example `.supercoder.yaml`:**
92+
**Example `~/.supercoder/config.yaml`:**
9393
```yaml
94-
api_key: "sk-..."
95-
model: "gpt-4o"
96-
base_url: "https://api.openai.com/v1"
97-
max_context_tokens: 32000
98-
use_repo_map: true
99-
100-
# Multiple model profiles
101-
model_profiles:
102-
gpt4:
103-
model: "gpt-4o"
104-
base_url: "https://api.openai.com/v1"
105-
local:
106-
model: "qwen2.5-coder:7b"
107-
base_url: "http://localhost:11434/v1"
94+
# Default model profile to use on startup
95+
default_model: "default"
96+
97+
# Model profiles - define multiple LLM configurations
98+
models:
99+
default:
100+
api_key: "sk-..."
101+
endpoint: "https://api.openai.com/v1"
102+
model: "gpt-4o-mini"
103+
104+
# OpenRouter with Qwen-style model
105+
openrouter-qwen:
106+
api_key: "sk-or-v1-..."
107+
endpoint: "https://openrouter.ai/api/v1"
108+
model: "openai/gpt-oss-20b:free"
109+
tool_calling_type: "qwen_like" # See Tool Calling Types below
110+
111+
# Local Ollama
112+
ollama:
108113
api_key: "ollama"
114+
endpoint: "http://localhost:11434/v1"
115+
model: "qwen2.5-coder:7b"
116+
tool_calling_type: "supercoder"
117+
118+
# Shared settings (applied to all models)
119+
temperature: 0.2
120+
max_context_tokens: 32000
121+
request_timeout: 60.0
122+
debug: false
109123
```
110124
125+
### Tool Calling Types
126+
127+
Different models expect tools to be called in different formats. Use `tool_calling_type` to specify the format:
128+
129+
| Type | Format | Best for |
130+
|------|--------|----------|
131+
| `supercoder` (default) | `<@TOOL>{"name": "...", "arguments": {...}}</@TOOL>` | Most instruction-following models |
132+
| `qwen_like` | `to=tool:name {"arg": "value"}` | Qwen, GPT-OSS, DeepResearch models |
133+
| `json_block` | ` ```json {"tool": "...", "arguments": {...}} ``` ` | Models trained on markdown |
134+
| `xml_function` | `<function_call name="...">...</function_call>` | XML-style models |
135+
111136
---
112137

113138
## ⌨️ Usage

0 commit comments

Comments
 (0)