Plugins

Plugins extend Codumentor with additional capabilities such as web search, persistent memory, performance metrics, and integrations with external tools. Codumentor ships with a set of built-in plugins that you can enable in your configuration.

Built-in Plugins

The table below lists the plugins that ship with Codumentor. Add one by its
module path (all modules live under codumentor.plugins.) — e.g. module
telemetry is codumentor.plugins.telemetry. See
Enabling and Disabling Plugins below.

Plugin	Module	Purpose	Docs
Agentic Memory	`agenticmemory`	Store agent responses as searchable organizational knowledge; retrieve during chats; web review UI	Details
Auto Title	`auto_title`	Name conversations from their content; opt-in topic-shift retitling	Details
Branch Switching	`branch_switching`	Per-conversation git branch selection with isolated checkout and branch-specific index	Details
Bug Report	`bug_report`	User bug/feedback form forwarded to Redmine as tracked issues	Details
Claude Code Hooks	`claude_code_hooks`	Run shell/HTTP callouts on lifecycle events (Claude Code-compatible)	Details
Client Filesystem	`client_fs`	Browser-attached local folder access (agent read/write, @-mention)	Details
Configuration Assistant	`configuration_assistant`	Admin chat that edits this instance's own config (validate → diff → apply → rollback)	Details
Console Report	`console_report`	End-of-session report renderer	Details
Context Management	`context_management`	Proactive multi-resolution summarization with semantic drift detection	Details
Context Summarization	`context_summarization`	Context-window management; truncates long conversations intelligently	Details
Database	`database`	SQL tools over switchable native/CLI backends (Postgres/MySQL/Oracle/SQL Server)	Details
Developer	`developer`	Subagent that reads, edits, and runs code from a plan	Details
Document Editor	`doc_editor`	Read, edit, and save DOCX documents	Details
EULA	`eula`	First-login End-User License Agreement modal	Details
Explore	`explore`	Read-only fan-out code-exploration subagent	Details
External Agent	`external_agent`	Delegate to external CLI coding agents (Claude Code, Codex) in a sandbox	Details
File Links	`file_links`	Browse/open repository files referenced in chat; read-only share-file access	Details
FileSync	`filesync`	Sync repos from local/SMB/UNC paths before ingestion	Details
Goals	`goals`	Long-running goal workers with scheduled/agentic execution	Details
Inference Metrics	`inference_metrics`	LLM generation performance (TTFT, tokens/sec)	Details
LLM Server	`llm_server`	Auto-start, health-monitor, and idle-shutdown a self-hosted LLM server	Details
Log Collection	`log_collection`	Log-file analysis source type	Details
MCP	`mcp`	Connect external Model Context Protocol tool servers (stdio/HTTP/SSE)	Details
Mermaid Render	`mermaid_render`	Render Mermaid diagrams to images (requires Playwright + Chromium)	Details
Model Profiles	`model_profiles`	Per-user selectable model profiles and custom API keys	Details
OAuth	`oauth`	Per-user OAuth 2.0 / refresh-token storage and flows	Details
Permalink	`permalink`	Auto-link file paths in responses to GitHub/GitLab/Gitea	Details
PPTX Editor	`pptx_editor`	Read, edit, and save PPTX presentations	Details
Read Aloud	`read_aloud`	Text-to-speech "read aloud" message action	Details
Redmine	`redmine`	Per-user Redmine issue tools (search/get/create/update/comment)	Details
Remote SSH	`remote_ssh`	Remote SSH/SFTP command and file tools (experimental)	Details
Ripgrep	`ripgrep`	Fast regex code search across repositories	Details
Scheduler Tools	`scheduler_tools`	Agent-callable scheduled / cron task tools	Details
SCM Push	`scm_push`	Secure in-sandbox `git_commit` plus host-side push gateway	Details
Script Filter	`script_filter`	Strip leaked non-Latin (CJK) output live; translate removed fragments on finish	Details
Skills	`skills`	Agent Skills (agentskills.io) activated via slash commands	Details
Slack Integration	`slack_integration`	Inbound/outbound Slack channel (mentions, DMs, slash commands, post/react tools)	Details
Slash Commands	`slashcmds`	Expand user-defined `/commands` into full prompts	Details
SSH Tunnel	`ssh_tunnel`	Persistent SSH port-forward tunnel to a fixed remote server	Details
SVN	`svn`	Subversion repository support (checkout/update, branch listing/switch)	Details
Telegram	`telegram`	Telegram bot channel with voice in/out (STT/TTS)	Details
Telemetry	`telemetry`	Response timing and LLM/tool metrics	Details
Token Report	`token_report`	Token usage and cost section for the session report	Details
Trial	`trial`	Cosmetic trial badge in the chat header	Details
User Feedback	`user_feedback`	Turn-level like/dislike response feedback	Details
User Secrets	`user_secrets`	Per-user encrypted credentials, referenced from config as `${secret:<name>}`	Details
Web Retrieve	`webretrieve`	Fetch and extract web-page content (optional agentic mode)	Details
Web Search	`websearch`	Web search across multiple providers, with optional research mode	Details
Workspace Isolation	`workspace_isolation`	Per-conversation bwrap overlay sandbox with snapshots and subagent integration	Details
XLSX Editor	`xlsx_editor`	Read, edit, and save XLSX spreadsheets	Details

Enabling and Disabling Plugins

Add plugins to the plugins list in codumentor.yaml:

plugins:
  # Enable telemetry
  - module: "codumentor.plugins.telemetry"
    class: "TelemetryPlugin"

  # Enable web search
  - module: "codumentor.plugins.web_search"
    class: "WebSearchPlugin"

  # Enable agentic memory
  - module: "codumentor.plugins.agentic_memory"
    class: "AgenticMemoryPlugin"

To disable a plugin, remove it from the list. Most plugins also accept enabled: false in their args:

plugins:
  - module: "codumentor.plugins.web_search"
    class: "WebSearchPlugin"
    args:
      enabled: false   # Loaded but inactive

Plugin Categories

The notes below highlight a few commonly used plugins by category. For the
complete list see Built-in Plugins above, and for
per-plugin configuration see the Plugin Catalog.

Observability

Telemetry -- Tracks response wall time, LLM call durations, and tool usage metrics. Exports data for evaluation reports.
Inference Metrics -- Monitors LLM generation performance including Time to First Token (TTFT) and tokens per second, with rolling averages tiered by prompt size.
Console Report -- Assembles and prints a formatted session report on exit, combining sections from Telemetry, Inference Metrics, and other plugins.

Context Management

Context Summarization -- Automatically truncates long conversations while preserving important context, with configurable per-model limits.
Context Management -- Proactive multi-resolution summarization with cascading summaries at multiple token levels and LLM-based semantic drift detection.

Knowledge and Memory

Agentic Memory -- Stores agent responses as searchable organizational knowledge. Retrieves relevant knowledge during conversations. Includes a web UI for voting and review.

Tool Integration

MCP -- Connects to local (stdio) or remote (HTTP/SSE) Model Context Protocol servers, exposing their tools to the agent.
Ripgrep -- Fast regex-based code search across repositories, replacing shell-based grep commands.
Web Retrieve -- Fetches and extracts text from web pages, with an optional agentic mode for complex information extraction.
Web Search -- Searches the web using multiple providers (DuckDuckGo, Brave, Tavily, Serper, Google CSE, SearXNG) with an optional research mode.

Workspace Isolation

Workspace Isolation -- Per-conversation bwrap sandbox with overlay filesystem. Subagent changes are automatically integrated via VCS-aware diffing with AI conflict resolution. Stable snapshots decouple sessions from ingestion pulls.

Infrastructure

LLM Server -- Manages a self-hosted LLM server (e.g. vLLM). Auto-detects connection errors, starts the server on demand, monitors health, and shuts down after inactivity.

Extensibility

Claude Code Hooks -- Runs external shell commands or HTTP endpoints on agent lifecycle events using the same JSON-in/JSON-out protocol as Anthropic's Claude Code. Supports inline rules, external files, and opt-in .claude/settings.json discovery. Exposes six Claude Code events for drop-in compatibility plus seven Codumentor-unique events (PreLLMCall, PostLLMCall, LLMError, PromptAssemble, TurnInterrupted, ToolProgress, ResponsePersist).