mirror of
https://github.com/Palm1r/QodeAssist.git
synced 2026-07-08 14:09:14 -04:00
fix: Found and fix review mistakes
This commit is contained in:
Petr Mironychev
@@ -63,15 +63,24 @@ Agent(config, provider)
|
||||
validated at load against a synthetic context
|
||||
provider.setPromptCaching(cfg.cachePrompt, cfg.cacheTtl == "1h")
|
||||
▼
|
||||
SessionManager — two ways to obtain a Session:
|
||||
• createSession(agentName, externalHistory?) — chat: attaches a persistent,
|
||||
externally-owned history
|
||||
SessionManager — three ways to obtain a Session:
|
||||
• createDetachedSession(externalHistory) — chat: one persistent, caller-owned
|
||||
session over an externally-owned
|
||||
history; the agent is (re)bound via
|
||||
rebindAgentByName when the picker
|
||||
changes (rebinding cancels any
|
||||
in-flight request)
|
||||
• createSession(agentName, externalHistory?) — manager-tracked session, removed
|
||||
via removeSession
|
||||
• acquire(agentName) / release(session) — one-shot pipelines: a small
|
||||
per-agent pool of internal-history
|
||||
sessions; acquire hands out a
|
||||
session with cleared history,
|
||||
cleared system-prompt layers and
|
||||
cleared client tools
|
||||
cleared system-prompt layers,
|
||||
cleared client tools and cleared
|
||||
pinned context / content loader /
|
||||
bindings; release refuses to pool
|
||||
external-history sessions
|
||||
▼
|
||||
Session(agent[, externalHistory])
|
||||
├─ ConversationHistory — messages as polymorphic ContentBlocks
|
||||
@@ -214,20 +223,24 @@ ChatRootView (QML) — owns ConversationHistory m_history
|
||||
chatAssistant allow-list; active agent persisted
|
||||
▼ dispatchSend
|
||||
ClientInterface
|
||||
session = sessionManager.createSession(activeAgent, m_history)
|
||||
session = sessionManager.createDetachedSession(m_history) — one persistent
|
||||
session per chat view; ensureAgentBound → rebindAgentByName on
|
||||
picker change
|
||||
sessionManager.toolContributors().contribute(client.tools()) — builtin+skills+MCP
|
||||
session.setContentLoader(ChatSerializer::loadContentFromStorage)
|
||||
systemPrompt layer "chat.context" = project info + skills + linked files
|
||||
session.send( blocks{ TextContent + StoredAttachmentContent + StoredImageContent } )
|
||||
▼ consumes Session signals (NOT raw client signals):
|
||||
event(Usage) → ChatModel.setMessageUsage + token-counter calibration
|
||||
finished(id) → ChangesManager.applyPendingEditsForRequest + persist;
|
||||
removeSession (the persistent history survives)
|
||||
failed(id, ErrorInfo) → surface error; removeSession
|
||||
finished(id) → ChangesManager.applyPendingEditsForRequest + persist
|
||||
failed(id, ErrorInfo) → surface error
|
||||
cancelled(id) → clear busy state (the persistent session and history survive
|
||||
every outcome)
|
||||
|
||||
ChatCompressor → acquire(chatCompression agent — single configured) → seed history
|
||||
from the chat's messages → "compression" layer → send → read summary
|
||||
from the compression session's own history → release
|
||||
ChatCompressor → acquire(chatCompression agent — single configured) → send the
|
||||
flattened transcript as one user message → read summary from the
|
||||
compression session's own history → write a fresh v0.3 file →
|
||||
release
|
||||
InputTokenCounter → estimates over ConversationHistory (calibrated by Usage events)
|
||||
ChatSerializer → persists ConversationHistory via MessageSerializer (v0.3);
|
||||
imports legacy v0.1/v0.2 files
|
||||
|
||||
@@ -1,16 +1,12 @@
|
||||
# Configure for Anthropic Claude
|
||||
|
||||
1. Open Qt Creator settings and navigate to the QodeAssist section
|
||||
2. Go to Provider Settings tab and configure Claude api key
|
||||
3. Return to General tab and configure:
|
||||
- Set "Claude" as the provider for code completion or/and chat assistant
|
||||
- Set the Claude URL (https://api.anthropic.com)
|
||||
- Select your preferred model (e.g., claude-3-5-sonnet-20241022)
|
||||
- Choose the Claude template for code completion or/and chat
|
||||
|
||||
<details>
|
||||
<summary>Example of Claude settings: (click to expand)</summary>
|
||||
|
||||
<img width="823" alt="Claude Settings" src="https://github.com/user-attachments/assets/828e09ea-e271-4a7a-8271-d3d5dd5c13fd" />
|
||||
</details>
|
||||
1. Open Qt Creator settings and navigate to QodeAssist > Providers
|
||||
2. Select the bundled **Claude** provider and enter your Anthropic API key
|
||||
3. Go to QodeAssist > General > Agent Pipelines and assign Claude agents to the features you want:
|
||||
- Code completion: **Claude Completion**
|
||||
- Chat assistant: **Claude Chat — Sonnet** (or an Opus variant)
|
||||
- Chat compression: **Claude Compression**
|
||||
- Quick refactor: **Claude Quick Refactor** (or the Fast variant)
|
||||
|
||||
To change the model or request parameters, duplicate the bundled agent on the
|
||||
QodeAssist > Agents page (it extends the Claude base agent) and edit your copy.
|
||||
|
||||
@@ -1,16 +1,12 @@
|
||||
# Configure for Google AI
|
||||
|
||||
1. Open Qt Creator settings and navigate to the QodeAssist section
|
||||
2. Go to Provider Settings tab and configure Google AI api key
|
||||
3. Return to General tab and configure:
|
||||
- Set "Google AI" as the provider for code completion or/and chat assistant
|
||||
- Set the OpenAI URL (https://generativelanguage.googleapis.com/v1beta)
|
||||
- Select your preferred model (e.g., gemini-2.0-flash)
|
||||
- Choose the Google AI template
|
||||
|
||||
<details>
|
||||
<summary>Example of Google AI settings: (click to expand)</summary>
|
||||
|
||||
<img width="829" alt="Google AI Settings" src="https://github.com/user-attachments/assets/046ede65-a94d-496c-bc6c-41f3750be12a" />
|
||||
</details>
|
||||
1. Open Qt Creator settings and navigate to QodeAssist > Providers
|
||||
2. Select the bundled **Google AI** provider and enter your Google AI Studio API key
|
||||
3. Go to QodeAssist > General > Agent Pipelines and assign Google agents to the features you want:
|
||||
- Code completion: **Google Completion**
|
||||
- Chat assistant: **Google Chat**
|
||||
- Chat compression: **Google Compression**
|
||||
- Quick refactor: **Google Quick Refactor**
|
||||
|
||||
To change the Gemini model or request parameters, duplicate the bundled agent on
|
||||
the QodeAssist > Agents page (it extends the Google base agent) and edit your copy.
|
||||
|
||||
@@ -1,16 +1,14 @@
|
||||
# Configure for llama.cpp
|
||||
|
||||
1. Open Qt Creator settings and navigate to the QodeAssist section
|
||||
2. Go to General tab and configure:
|
||||
- Set "llama.cpp" as the provider for code completion or/and chat assistant
|
||||
- Set the llama.cpp URL (e.g. http://localhost:8080)
|
||||
- Fill in model name
|
||||
- Choose template for model(e.g. llama.cpp FIM for any model with FIM support)
|
||||
- Disable using tools if your model doesn't support tooling
|
||||
|
||||
<details>
|
||||
<summary>Example of llama.cpp settings: (click to expand)</summary>
|
||||
|
||||
<img width="829" alt="llama.cpp Settings" src="https://github.com/user-attachments/assets/8c75602c-60f3-49ed-a7a9-d3c972061ea2" />
|
||||
</details>
|
||||
1. Start `llama-server` locally (default http://localhost:8080)
|
||||
2. Open Qt Creator settings and navigate to QodeAssist > Providers
|
||||
- Select the bundled **llama.cpp** provider and adjust the URL if your server runs elsewhere (no API key needed)
|
||||
3. Go to QodeAssist > General > Agent Pipelines and assign llama.cpp agents to the features you want:
|
||||
- Code completion: **LlamaCpp Completion — FIM** (needs a FIM-capable model)
|
||||
- Chat assistant: **LlamaCpp Chat**
|
||||
- Chat compression: **LlamaCpp Compression**
|
||||
- Quick refactor: **LlamaCpp Quick Refactor**
|
||||
|
||||
To change the model or request parameters, duplicate the bundled agent on the
|
||||
QodeAssist > Agents page (it extends the llama.cpp base agent) and edit your
|
||||
copy. Disable `enable_tools` in your agent if the model doesn't support tool calling.
|
||||
|
||||
@@ -1,16 +1,13 @@
|
||||
# Configure for Mistral AI
|
||||
|
||||
1. Open Qt Creator settings and navigate to the QodeAssist section
|
||||
2. Go to Provider Settings tab and configure Mistral AI api key
|
||||
3. Return to General tab and configure:
|
||||
- Set "Mistral AI" as the provider for code completion or/and chat assistant
|
||||
- Set the OpenAI URL (https://api.mistral.ai)
|
||||
- Select your preferred model (e.g., mistral-large-latest)
|
||||
- Choose the Mistral AI template for code completion or/and chat
|
||||
|
||||
<details>
|
||||
<summary>Example of Mistral AI settings: (click to expand)</summary>
|
||||
|
||||
<img width="829" alt="Mistral AI Settings" src="https://github.com/user-attachments/assets/1c5ed13b-a29b-43f7-b33f-2e05fdea540c" />
|
||||
</details>
|
||||
1. Open Qt Creator settings and navigate to QodeAssist > Providers
|
||||
2. Select the bundled **Mistral AI** provider and enter your Mistral API key
|
||||
- For Codestral code completion, configure the separate **Codestral** provider — it uses its own key and endpoint (codestral.mistral.ai)
|
||||
3. Go to QodeAssist > General > Agent Pipelines and assign Mistral agents to the features you want:
|
||||
- Code completion: **Mistral Completion — Codestral FIM** (needs the Codestral provider)
|
||||
- Chat assistant: **Mistral Chat** (or the Reasoning variant)
|
||||
- Chat compression: **Mistral Compression**
|
||||
- Quick refactor: **Mistral Quick Refactor**
|
||||
|
||||
To change the model or request parameters, duplicate the bundled agent on the
|
||||
QodeAssist > Agents page (it extends the Mistral base agent) and edit your copy.
|
||||
|
||||
@@ -17,14 +17,13 @@ ollama run qwen2.5-coder:32b
|
||||
```
|
||||
|
||||
3. Open Qt Creator settings (Edit > Preferences on Linux/Windows, Qt Creator > Preferences on macOS)
|
||||
4. Navigate to the "QodeAssist" tab
|
||||
5. On the "General" page, verify:
|
||||
- Ollama is selected as your LLM provider
|
||||
- The URL is set to http://localhost:11434
|
||||
- Your installed model appears in the model selection
|
||||
- The prompt template is Ollama Auto FIM or Ollama Auto Chat for chat assistance. You can specify template if it is not work correct
|
||||
- Disable using tools if your model doesn't support tooling
|
||||
6. Click Apply if you made any changes
|
||||
4. Navigate to QodeAssist > Providers and verify the bundled **Ollama (Native)** provider points at http://localhost:11434 (no API key needed)
|
||||
5. Navigate to QodeAssist > General > Agent Pipelines — the Ollama agents are assigned by default:
|
||||
- Code completion: **Ollama Completion — FIM** (needs a base/FIM model; use **Ollama Completion — Chat-style** for instruct models)
|
||||
- Chat assistant: **Ollama Chat — Simple / Thinking / Gemma 4**
|
||||
- Chat compression: **Ollama Compression — 8 GB** (or the 16/32 GB tier for your machine)
|
||||
- Quick refactor: **Ollama Quick Refactor — Simple**
|
||||
6. Point the agents at models you actually have (see the next section)
|
||||
|
||||
You're all set! QodeAssist is now ready to use in Qt Creator.
|
||||
|
||||
@@ -55,45 +54,13 @@ swap it (Change…) for one you already have.
|
||||
|
||||
## Extended Thinking Mode
|
||||
|
||||
Ollama supports extended thinking mode for models that are capable of deep reasoning (such as DeepSeek-R1, QwQ, and similar reasoning models). This mode allows the model to show its step-by-step reasoning process before providing the final answer.
|
||||
Ollama supports native reasoning for models trained for it (Qwen3.5, Gemma 4, DeepSeek-R1, QwQ, …). Reasoning is streamed into collapsible "Thinking" blocks in the chat.
|
||||
|
||||
### How to Enable
|
||||
Thinking is a property of the **agent**, not a global switch:
|
||||
|
||||
**For Chat Assistant:**
|
||||
1. Navigate to Qt Creator > Preferences > QodeAssist > Chat Assistant
|
||||
2. In the "Extended Thinking (Claude, Ollama)" section, check "Enable extended thinking mode"
|
||||
3. Select a reasoning-capable model (e.g., deepseek-r1:8b, qwq:32b)
|
||||
4. Click Apply
|
||||
- For chat, pick a thinking agent in the chat panel: **Ollama Chat — Thinking** or **Ollama Chat — Gemma 4**
|
||||
- For quick refactor, assign **Ollama Quick Refactor — Qwen3.5** or **— Gemma 4** in Agent Pipelines
|
||||
- In your own agents, set `think = true` in the `[body]` table (top level, not under `[body.options]`)
|
||||
|
||||
**For Quick Refactoring:**
|
||||
1. Navigate to Qt Creator > Preferences > QodeAssist > Quick Refactor
|
||||
2. Check "Enable Thinking Mode"
|
||||
3. Configure thinking budget and max tokens as needed
|
||||
4. Click Apply
|
||||
|
||||
### Supported Models
|
||||
|
||||
Thinking mode works best with models specifically designed for reasoning:
|
||||
- **DeepSeek-R1** series (deepseek-r1:8b, deepseek-r1:14b, deepseek-r1:32b)
|
||||
- **QwQ** series (qwq:32b)
|
||||
- Other models trained for chain-of-thought reasoning
|
||||
|
||||
### How It Works
|
||||
|
||||
When thinking mode is enabled:
|
||||
1. The model generates internal reasoning (visible in the chat as "Thinking" blocks)
|
||||
2. After reasoning, it provides the final answer
|
||||
3. You can collapse/expand thinking blocks to focus on the final answer
|
||||
4. Temperature is automatically set to 1.0 for optimal reasoning performance
|
||||
|
||||
**Technical Details:**
|
||||
- Thinking mode adds the `enable_thinking: true` parameter to requests sent to Ollama
|
||||
- This is natively supported by the Ollama API for compatible models
|
||||
- Works in both Chat Assistant and Quick Refactoring contexts
|
||||
|
||||
<details>
|
||||
<summary>Example of Ollama settings: (click to expand)</summary>
|
||||
|
||||
<img width="824" alt="Ollama Settings" src="https://github.com/user-attachments/assets/ed64e03a-a923-467a-aa44-4f790e315b53" />
|
||||
</details>
|
||||
Use a reasoning-capable model with these agents — a non-reasoning model simply ignores the flag.
|
||||
|
||||
|
||||
@@ -1,32 +1,18 @@
|
||||
# Configure for OpenAI
|
||||
|
||||
QodeAssist supports both OpenAI's standard Chat Completions API and the new Responses API, giving you access to the latest GPT models including GPT-5.1 and GPT-5.1-codex.
|
||||
QodeAssist supports both OpenAI's standard Chat Completions API and the Responses API, giving you access to the latest GPT models.
|
||||
|
||||
## Standard OpenAI Configuration
|
||||
1. Open Qt Creator settings and navigate to QodeAssist > Providers
|
||||
2. Select the bundled **OpenAI (Chat Completions)** or **OpenAI (Responses API)** provider and enter your OpenAI API key
|
||||
3. Go to QodeAssist > General > Agent Pipelines and assign OpenAI agents to the features you want:
|
||||
- Code completion: **OpenAI Completion**
|
||||
- Chat assistant: **OpenAI Chat**, **OpenAI Chat — Mini** (Chat Completions), or **OpenAI Chat — Responses**
|
||||
- Chat compression: **OpenAI Compression**
|
||||
- Quick refactor: **OpenAI Quick Refactor**
|
||||
|
||||
1. Open Qt Creator settings and navigate to the QodeAssist section
|
||||
2. Go to Provider Settings tab and configure OpenAI api key
|
||||
3. Return to General tab and configure:
|
||||
- Set "OpenAI" as the provider for code completion or/and chat assistant
|
||||
- Set the OpenAI URL (https://api.openai.com)
|
||||
- Select your preferred model (e.g., gpt-4o, gpt-5.1, gpt-5.1-codex)
|
||||
- Choose the OpenAI template for code completion or/and chat
|
||||
|
||||
<details>
|
||||
<summary>Example of OpenAI settings: (click to expand)</summary>
|
||||
|
||||
<img width="829" alt="OpenAI Settings" src="https://github.com/user-attachments/assets/4716f790-6159-44d0-a8f4-565ccb6eb713" />
|
||||
</details>
|
||||
|
||||
## OpenAI Responses API Configuration
|
||||
|
||||
The Responses API is OpenAI's newer endpoint that provides enhanced capabilities and improved performance. It supports the latest GPT-5.1 models.
|
||||
|
||||
1. Open Qt Creator settings and navigate to the QodeAssist section
|
||||
2. Go to Provider Settings tab and configure OpenAI api key
|
||||
3. Return to General tab and configure:
|
||||
- Set "OpenAI Responses" as the provider for code completion or/and chat assistant
|
||||
- Set the OpenAI URL (https://api.openai.com)
|
||||
- Select your preferred model (e.g., gpt-5.1, gpt-5.1-codex)
|
||||
- Choose the OpenAI Responses template for code completion or/and chat
|
||||
To change the model or request parameters, duplicate the bundled agent on the
|
||||
QodeAssist > Agents page (it extends the OpenAI base agent) and edit your copy.
|
||||
|
||||
Note for GPT-5 family models on Chat Completions: use `max_completion_tokens`
|
||||
(not `max_tokens`) and `reasoning_effort` in the agent `[body]`; reasoning
|
||||
models reject `temperature`.
|
||||
|
||||
@@ -9,11 +9,11 @@ Make sure you're using the correct default URLs:
|
||||
- **LM Studio**: `http://localhost:1234`
|
||||
- **llama.cpp**: `http://localhost:8080`
|
||||
|
||||
### 2. Check model and template compatibility
|
||||
### 2. Check agent and model compatibility
|
||||
|
||||
- Ensure the correct model is selected in settings
|
||||
- Verify that the selected prompt template matches your model
|
||||
- Some models may not support certain features (e.g., tool calling)
|
||||
- Ensure each Agent Pipelines slot (QodeAssist > General) has an agent assigned — an unassigned slot disables that feature
|
||||
- Verify the agent's model exists on your provider (check the agent on the QodeAssist > Agents page)
|
||||
- Some models may not support certain features (e.g., tool calling or native FIM completion) — pick an agent variant that matches your model
|
||||
|
||||
### 3. Linux compatibility
|
||||
|
||||
@@ -31,7 +31,7 @@ If issues persist, you can reset settings to their default values:
|
||||
|
||||
**Note:**
|
||||
- API keys are preserved during reset
|
||||
- You will need to re-select your model after reset
|
||||
- Resetting the General page restores the default local Ollama agent pipelines — re-assign your own agents afterwards
|
||||
|
||||
## Chat History Migration
|
||||
|
||||
|
||||
Reference in New Issue
Block a user