内容
# CodeCompanion History Extension
[](https://neovim.io)
[](https://www.lua.org)
[](https://github.com/ravitemer/codecompanion-history.nvim/actions)
[](https://opensource.org/licenses/MIT)
[](./CONTRIBUTING.md)
A history management extension for [codecompanion.nvim](https://codecompanion.olimorris.dev/) that enables saving, browsing and restoring chat sessions.
<p>
<video controls muted src="https://github.com/user-attachments/assets/04a6ad1f-8351-4381-ae60-00c352a1670c"></video>
</p>
## ✨ Features
### 🤖 Chat Management
- 💾 Flexible chat saving:
- Automatic session saving (can be disabled)
- Manual save with dedicated keymap
- 🎯 Smart title generation for chats
- 🔄 Continue from where you left
- 📚 Browse saved chats with preview
- 🔍 Multiple picker interfaces
- ⌛ Optional automatic chat expiration
- ⚡ Restore chat sessions with full context and tools state
- 🏢 **Project-aware filtering**: Filter chats by workspace/project context
- 📋 **Chat duplication**: Easily duplicate chats to create variations or backups
### 📝 Summary System
- **Manual summary generation**: Create summaries for any chat with `gcs`
- **Intelligent content processing**: Extracts meaningful conversation content while filtering noise
- **Chunked summarization**: Handles large conversations by splitting into manageable chunks
- **Customizable generation**: Configure adapter, model, and system prompts
- **Summary browsing**: Dedicated browser with `gbs` to explore all summaries
### 🧠 Memory System (@memory tool)
- **Vector-based search**: Uses VectorCode CLI to index and search through chat summaries
- **Automatic indexing**: Optionally index summaries as they are generated
- **Smart integration**: Available as `@memory` tool in new chats when VectorCode is installed
The following CodeCompanion features are preserved when saving and restoring chats:
| Feature | Status | Notes |
|---------|--------|-------|
| System Prompts | ✅ | System prompt used in the chat |
| Messages History | ✅ | All messages |
| Images | ✅ | Restores images as base64 strings |
| LLM Adapter | ✅ | The specific adapter used for the chat |
| LLM Settings | ✅ | Model, temperature and other adapter settings |
| Tools | ✅ | Tool schemas and their system prompts |
| Tool Outputs | ✅ | Tool execution results |
| Variables | ✅ | Variables used in the chat |
| References | ✅ | Code snippets and command outputs added via slash commands |
| Pinned References | ✅ | Pinned references |
| Watchers | ⚠ | Saved but requires original buffer context to resume watching |
When restoring a chat:
1. The complete message history is recreated
2. All tools and references are reinitialized
3. Original LLM settings and adapter are restored
4. Previous system prompts are preserved
> **Note**: While watched buffer states are saved, they require the original buffer context to resume watching functionality.
> [!NOTE]
> As this is an extension that deeply integrates with CodeCompanion's internal APIs, occasional compatibility issues may arise when CodeCompanion updates. If you encounter any bugs or unexpected behavior, please [raise an issue](https://github.com/ravitemer/codecompanion-history.nvim/issues) to help us maintain compatibility.
## 📋 Requirements
- Neovim >= 0.8.0
- [codecompanion.nvim](https://codecompanion.olimorris.dev/)
- [VectorCode CLI](https://github.com/Davidyz/VectorCode) (optional, for `@memory` tool)
- [snacks.nvim](https://github.com/folke/snacks.nvim) (optional, for enhanced picker)
- [telescope.nvim](https://github.com/nvim-telescope/telescope.nvim) (optional, for enhanced picker)
- [fzf-lua](https://github.com/ibhagwan/fzf-lua) (optional, for enhanced picker)
## 📦 Installation
Using [lazy.nvim](https://github.com/folke/lazy.nvim):
### First install the plugin
```lua
{
"olimorris/codecompanion.nvim",
dependencies = {
--other plugins
"ravitemer/codecompanion-history.nvim"
}
}
```
### Add history extension to CodeCompanion config
```lua
require("codecompanion").setup({
extensions = {
history = {
enabled = true,
opts = {
-- Keymap to open history from chat buffer (default: gh)
keymap = "gh",
-- Keymap to save the current chat manually (when auto_save is disabled)
save_chat_keymap = "sc",
-- Save all chats by default (disable to save only manually using 'sc')
auto_save = true,
-- Number of days after which chats are automatically deleted (0 to disable)
expiration_days = 0,
-- Picker interface (auto resolved to a valid picker)
picker = "telescope", --- ("telescope", "snacks", "fzf-lua", or "default")
---Optional filter function to control which chats are shown when browsing
chat_filter = nil, -- function(chat_data) return boolean end
-- Customize picker keymaps (optional)
picker_keymaps = {
rename = { n = "r", i = "<M-r>" },
delete = { n = "d", i = "<M-d>" },
duplicate = { n = "<C-y>", i = "<C-y>" },
},
---Automatically generate titles for new chats
auto_generate_title = true,
title_generation_opts = {
---Adapter for generating titles (defaults to current chat adapter)
adapter = nil, -- "copilot"
---Model for generating titles (defaults to current chat model)
model = nil, -- "gpt-4o"
---Number of user prompts after which to refresh the title (0 to disable)
refresh_every_n_prompts = 0, -- e.g., 3 to refresh after every 3rd user prompt
---Maximum number of times to refresh the title (default: 3)
max_refreshes = 3,
format_title = function(original_title)
-- this can be a custom function that applies some custom
-- formatting to the title.
return original_title
end
},
---On exiting and entering neovim, loads the last chat on opening chat
continue_last_chat = false,
---When chat is cleared with `gx` delete the chat from history
delete_on_clearing_chat = false,
---Directory path to save the chats
dir_to_save = vim.fn.stdpath("data") .. "/codecompanion-history",
---Enable detailed logging for history extension
enable_logging = false,
-- Summary system
summary = {
-- Keymap to generate summary for current chat (default: "gcs")
create_summary_keymap = "gcs",
-- Keymap to browse summaries (default: "gbs")
browse_summaries_keymap = "gbs",
generation_opts = {
adapter = nil, -- defaults to current chat adapter
model = nil, -- defaults to current chat model
context_size = 90000, -- max tokens that the model supports
include_references = true, -- include slash command content
include_tool_outputs = true, -- include tool execution results
system_prompt = nil, -- custom system prompt (string or function)
format_summary = nil, -- custom function to format generated summary e.g to remove <think/> tags from summary
},
},
-- Memory system (requires VectorCode CLI)
memory = {
-- Automatically index summaries when they are generated
auto_create_memories_on_summary_generation = true,
-- Path to the VectorCode executable
vectorcode_exe = "vectorcode",
-- Tool configuration
tool_opts = {
-- Default number of memories to retrieve
default_num = 10
},
-- Enable notifications for indexing progress
notify = true,
-- Index all existing memories on startup
-- (requires VectorCode 0.6.12+ for efficient incremental indexing)
index_on_startup = false,
},
}
}
}
})
```
> [!WARNING]
> Title and summary generation defaults to current chat's adapter and model. Make sure to set cheaper models in `title_generation_opts` and `summary.generation_opts` to avoid using premium models.
## 🛠️ Usage
#### 🎯 Commands
- `:CodeCompanionHistory` - Open the history browser
- `:CodeCompanionSummaries` - Browse all summaries
#### ⌨️ Chat Buffer Keymaps
**History Management:**
- `gh` - Open history browser (customizable via `opts.keymap`)
- `sc` - Save current chat manually (customizable via `opts.save_chat_keymap`)
**Summary System:**
- `gcs` - Generate summary for current chat (customizable via `opts.summary.create_summary_keymap`)
- `gbs` - Browse saved summaries (customizable via `opts.summary.browse_summaries_keymap`)
#### 📚 History Browser
The history browser shows all your saved chats with:
- Title (auto-generated or custom)
- Summary indicator (📝 icon for chats with summaries)
- Token estimates and relative timestamps
- Preview of chat contents
Actions in history browser:
- `<CR>` - Open selected chat
- Normal mode:
- `d` - Delete selected chat(s)
- `r` - Rename selected chat
- `<C-y>` - Duplicate selected chat
- Insert mode:
- `<M-d>` (Alt+d) - Delete selected chat(s)
- `<M-r>` (Alt+r) - Rename selected chat
- `<C-y>` - Duplicate selected chat
#### 📝 Summary Browser
The summary browser shows all your generated summaries with:
- Chat title (from original conversation)
- Project context and relative timestamps
- Preview of summary content
Actions in summary browser:
- `<CR>` - Add the summary to the current chat
- Normal mode:
- `d` - Delete selected summary(s)
- Insert mode:
- `<M-d>` (Alt+d) - Delete selected summary(s)
## The `@memory` tool
If you have installed the [VectorCode](https://github.com/Davidyz/VectorCode) CLI,
this plugin will use VectorCode to create an index for your chat summaries and create
a tool called `@memory`. This tool gives the LLM the ability to search for
(the summary of) previous chats so that you can refer to them in a new chat.
Available options for the memory submodule:
```lua
opts.memory = {
auto_create_memories_on_summary_generation = true,
-- path to the `vectorcode` executable
vectorcode_exe = "vectorcode",
tool_opts = {
-- default number of memories to retrieve
default_num = 10
},
-- whether to enable notification
notify = true,
-- whether to automatically update the index of all existing memories on startup
-- (requires VectorCode 0.6.12+ for efficient incremental indexing)
index_on_startup = false,
}
```
#### 🔄 Title Refresh Feature
The extension can automatically refresh chat titles as conversations evolve:
- **`refresh_every_n_prompts`**: Set to refresh the title after every N user prompts (e.g., 3 means refresh after the 3rd, 6th, 9th user message)
- **`max_refreshes`**: Limits how many times a title can be refreshed to avoid excessive API calls
- When refreshing, the system considers recent conversation context (both user and assistant messages) and the original title
- Individual messages are truncated at 1000 characters with a `[truncated]` indicator
- Total conversation context is limited to 10,000 characters with a `[conversation truncated]` indicator
Example configuration for title refresh:
```lua
title_generation_opts = {
refresh_every_n_prompts = 3, -- Refresh after every 3rd user prompt
max_refreshes = 10, -- Allow up to 10 refreshes per chat
}
```
#### 🏢 Project-Aware Chat Filtering
The extension supports flexible chat filtering to help you focus on relevant conversations:
**Configurable Filtering:**
```lua
chat_filter = function(chat_data)
return chat_data.cwd == vim.fn.getcwd()
end
-- Recent chats only (last 7 days)
chat_filter = function(chat_data)
local seven_days_ago = os.time() - (7 * 24 * 60 * 60)
return chat_data.updated_at >= seven_days_ago
end
```
**Chat Index Data Structure:**
Each chat index entry (used in filtering) includes the following information:
```lua
-- ChatIndexData - lightweight metadata used for browsing and filtering
{
save_id = "1672531200", -- Unique chat identifier
title = "Debug API endpoint", -- Chat title (auto-generated or custom)
cwd = "/home/user/my-project", -- Working directory when saved
project_root = "/home/user/my-project", -- Detected project root
adapter = "openai", -- LLM adapter used
model = "gpt-4", -- Model name
updated_at = 1672531200, -- Unix timestamp of last update
message_count = 15, -- Number of messages in chat
token_estimate = 3420, -- Estimated token count
}
```
#### 🔧 API
The history extension exports the following functions that can be accessed via `require("codecompanion").extensions.history`:
```lua
-- Chat Management
get_location(): string? -- Get storage location
-- Save a chat to storage (uses last chat if none provided)
save_chat(chat?: CodeCompanion.Chat)
-- Browse chats with custom filter function
browse_chats(filter_fn?: function(ChatIndexData): boolean)
-- Get metadata for all saved chats with optional filtering
get_chats(filter_fn?: function(ChatIndexData): boolean): table<string, ChatIndexData>
-- Load a specific chat by its save_id
load_chat(save_id: string): ChatData?
-- Delete a chat by its save_id
delete_chat(save_id: string): boolean
-- Duplicate a chat by its save_id
duplicate_chat(save_id: string, new_title?: string): string?
-- Summary Management
--- Generate a summary for the current chat
generate_summary(chat?: CodeCompanion.Chat)
--- Delete a sumamry
delete_summary(summary_id: string)
--- Get summaries index
get_summaries(): table<string, SummaryIndexData>
--- Load summary
load_summary(summary_id: string): string?
```
Example usage:
```lua
local history = require("codecompanion").extensions.history
-- Browse chats with project filter
history.browse_chats(function(chat_data)
return chat_data.project_root == utils.find_project_root()
end)
-- Get all saved chats metadata
local chats = history.get_chats()
local chat_data = history.load_chat("some_save_id")
history.delete_chat("some_save_id")
-- Duplicate a chat with custom title
local new_save_id = history.duplicate_chat("some_save_id", "My Custom Copy")
-- Duplicate a chat with auto-generated title (appends "(1)")
local new_save_id = history.duplicate_chat("some_save_id")
-- Summary operations
history.generate_summary() -- generates for current chat
local summaries = history.get_summaries()
local summary_content = history.load_summary("some_save_id")
```
## ⚙️ How It Works
```mermaid
graph TD
subgraph CodeCompanion Core Lifecycle
A[CodeCompanionChatCreated Event] --> B{Chat Submitted};
B --> C[LLM Response Received];
subgraph Chat End
direction RL
D[CodeCompanionChatCleared Event];
end
C --> D;
B --> D;
end
subgraph Extension Integration
A -- Extension Hooks --> E[Init & Subscribe];
E --> F[Setup Auto-Save];
F --> G[Prepare Auto-Title];
C -- Extension Hooks --> H[Subscriber Triggered];
H --> H1{Auto-Save Enabled?};
H1 -- Yes --> I[Save Chat State - Messages, Tools, Refs];
H1 -- No --> H2[Manual Save via `sc`];
H2 --> I;
I --> J{No Title & Auto-Title Enabled?};
J -- Yes --> K[Generate Title];
K --> L[Update Buffer Title];
L --> M[Save Chat with New Title];
J -- No --> B;
M --> B;
D -- Extension Hooks --> N[Respond to Clear Event];
N --> O[Delete Chat from Storage];
O --> P[Reset Extension State - Title/ID];
end
subgraph User History Interaction
Q[User Action - gh / :CodeCompanionHistory] --> R{History Browser};
R -- Restore --> S[Load Chat State from Storage];
S --> A;
R -- Delete --> O;
end
```
Here's what's happening in simple terms:
1. When you create a new chat, our extension jumps in and sets up two things:
- An autosave system that will save your chat
- A title generator that will name your chat based on the conversation
2. As you chat:
- When auto-save is enabled (default):
- Each submitted message triggers automatic saving
- Every LLM response automatically saves the chat state
- Manual saving is available via the `sc` keymap
- If your chat doesn't have a title yet, it tries to create one that makes sense
- All your messages, tools, and references are safely stored
3. When you clear a chat:
- Our extension knows to remove it from storage (if configured)
- This keeps your history clean and organized
4. Any time you want to look at old chats:
- Use `gh` or the command to open the history browser
- Pick any chat to restore it completely
- Or remove ones you don't need anymore
<details>
<summary> Technical details </summary>
The extension integrates with CodeCompanion through a robust event-driven architecture:
1. **Initialization and Storage Management**:
- Uses a dedicated Storage class to manage chat persistence in `{data_path}/codecompanion-history/`
- Maintains an index.json for metadata and individual JSON files for each chat
- Implements file I/O operations with error handling and atomic writes
2. **Chat Lifecycle Integration**:
- Hooks into `CodeCompanionChatCreated` event to:
- Generate unique save_id (Unix timestamp)
- Initialize chat subscribers for auto-saving
- Set initial buffer title with sparkle icon (✨)
- Monitors `CodeCompanionChatSubmitted` events to:
- Persist complete chat state including messages, tools, schemas, and references
- Trigger title generation if enabled and title is empty
- Update buffer title with relative timestamps
3. **Title Generation System**:
- Uses the chat's configured LLM adapter for title generation
- Implements smart content truncation (1000 chars) and prompt engineering
- Handles title collisions with automatic numbering
- Updates titles asynchronously using vim.schedule
4. **State Management**:
- Preserves complete chat context including:
- Message history with role-based organization
- Tool states and schemas
- Reference management
- Adapter configurations
- Custom settings
5. **UI Components**:
- Implements multiple picker interfaces (telescope/snacks/default)
- Provides real-time preview generation with markdown formatting
- Supports justified text layout for buffer titles
- Handles window/buffer lifecycle management
6. **Data Flow**:
- Chat data follows a structured schema (ChatData)
- Implements proper serialization/deserialization
- Maintains backward compatibility with existing chats
- Provides error handling for corrupt or missing data
</details>
## 🔮 Future Roadmap
### Upcoming Features
- [ ] Auto-summary generation options
- [ ] Summary search and filtering
- [ ] Integration with vector databases
## 🔌 Related Extensions
- [MCP Hub](https://codecompanion.olimorris.dev/extensions/mcphub.html) extension
- [VectorCode](https://codecompanion.olimorris.dev/extensions/vectorcode.html) extension
## 🙏 Acknowledgements
Special thanks to:
- [Oli Morris](https://github.com/olimorris) for creating the amazing [CodeCompanion.nvim](https://codecompanion.olimorris.dev) plugin - a highly configurable and powerful coding assistant for Neovim.
- [David](https://github.com/Davidyz) for the awesome [VectorCode](https://github.com/Davidyz/VectorCode) CLI and adding the @memory tool integration.
## 📄 License
MIT
连接信息
猜你喜欢
暂无推荐服务器