An always-on AI coding assistant for VS Code, built for your own self-hosted or OpenAI-compatible models.
Ollama, LiteLLM, vLLM, or any OpenAI-compatible endpoint — no vendor lock-in, no forced cloud dependency.
Why CodeOn
Most AI coding assistants tie you to one vendor's cloud API. CodeOn talks to any OpenAI-compatible chat completions endpoint — point it at a model you run yourself (Ollama on your laptop, a LiteLLM or vLLM gateway on your team's infra, or any hosted provider that speaks the same protocol) and get the same tool-calling, multi-file editing, and agentic workflow you'd expect from a cloud-only assistant, without your code ever having to leave infrastructure you control.
Features
- 5 chat modes — Plan, Build, Code Review, Debug, Research — each with a tailored system prompt for the task at hand
- Full tool system — file read/write/edit, git operations, process management, code intelligence (LSP, diagnostics, code actions), checkpoints (undo/redo across multi-step edits), and multi-agent orchestration for parallel sub-tasks
- MCP integration — connect to Model Context Protocol servers for additional tools and resources
- Inline edits & autocomplete — side-by-side diff preview for AI-proposed edits, plus ghost-text FIM autocomplete
- Policy engine — every tool call is risk-classified (R0–R5); anything beyond a safe read requires your approval by default, with configurable thresholds
- Persistent conversations — chats survive reloads, with automatic context summarization as conversations grow
- Codebase-aware retrieval — a lightweight local index (symbols, keywords, recency) feeds relevant context into requests automatically
- Audit trail — every tool invocation is logged locally with its risk class, outcome, and duration
Getting Started
1. Install
From the Marketplace — search "CodeOn AI" in the Extensions view (Ctrl+Shift+X / Cmd+Shift+X).
From source, for now:
git clone https://github.com/srathan1/codeon.git
cd codeon
npm install
npm run compile
Open the folder in VS Code and press F5 to launch an Extension Development Host with CodeOn loaded.
2. Connect a model
CodeOn needs at least one model provider configured before it can chat.
- Open the CodeOn panel from the Activity Bar
- Click the model dropdown at the top of the chat panel → Add Model
- Fill in:
- Provider name — anything you'll recognize (e.g. "Ollama", "Team LiteLLM")
- Endpoint — your OpenAI-compatible chat completions URL, e.g.
http://localhost:11434/v1 for a local Ollama instance, or your LiteLLM/vLLM gateway's URL
- API key — only if your endpoint requires one (local Ollama typically doesn't)
- Model name and its context window size
- Save, select the model from the dropdown, and start chatting
You can add multiple providers/models and switch between them at any time — useful for pairing a fast local model for autocomplete with a stronger model for the main chat.
3. Chat
Type a message and press Enter (Shift+Enter for a newline). Pick a mode from the dropdown depending on what you're doing — Plan for read-only exploration and design discussion, Build for making changes, Debug, Code Review, or Research. Attach files with the paperclip button when you need to point the model at something specific.
The first time a tool call needs to touch your filesystem, run a command, or do anything beyond reading, you'll be asked to approve it — see Configuration to adjust how eager that approval gate is.
Commands
Available via the Command Palette (Ctrl+Shift+P / Cmd+Shift+P):
| Command |
Description |
CodeOn: New Chat |
Start a fresh conversation |
CodeOn: Edit with Selection |
Send the current selection to chat for editing |
CodeOn: Accept Diff |
Apply a pending inline edit |
CodeOn: Reject Diff |
Discard a pending inline edit |
CodeOn: Show Request Logs |
View structured request logs |
CodeOn: Index Info |
Show codebase index statistics |
CodeOn: Rebuild Index |
Force-rebuild the symbol index |
CodeOn: Run Smoke Tests |
Execute the built-in smoke test suite |
CodeOn: Reset Conversation |
Clear the current chat's context |
Configuration
All settings live under "CodeOn" in VS Code Settings (Ctrl+, / Cmd+,). Model endpoint, API key, and context window are configured per-model via Add Model (see Getting Started), not here.
Sampling
| Setting |
Default |
Description |
temperature |
0.7 |
Sampling temperature (0 = deterministic, higher = more creative) |
topP |
0.8 |
Nucleus sampling top_p (0–1, lower = sharper focus) |
topK |
20 |
Top-k sampling (0 = disabled) |
presencePenalty |
1.5 |
Positive values encourage exploring new topics |
repetitionPenalty |
1 |
1.0 = neutral, higher suppresses repeated tokens |
minP |
0 |
Minimum probability floor for token selection (0 = no floor) |
Chat & interaction
| Setting |
Default |
Description |
defaultMode |
plan |
Default chat mode on startup |
openCodeEnabled |
true |
Enable tool calling |
openCodeApiKey |
(empty) |
OpenCode API key, if your setup requires one |
approvalThreshold |
moderate |
Risk level (safe / moderate / dangerous) that requires approval |
interactionMode |
ask |
ask (approve every tool), autoedit (auto-approve reads, prompt for edits), relaxed (only prompt for shell commands) |
assistantName |
(empty) |
Display name for the assistant; empty = use the model's name |
autoSummarizeThreshold |
75 |
Auto-summarize when context usage exceeds this % (0 disables) |
toolOutputMaxChars |
15000 |
Max characters per tool result kept in context |
Commands
| Setting |
Default |
Description |
commandTimeout |
30 |
Timeout (seconds) for execute_command |
commandBlocklist |
rm -rf, :(){ :|:};, mkfs, dd if= |
Patterns that block command execution outright |
Autocomplete
| Setting |
Default |
Description |
autocompleteEnabled |
true |
Enable ghost-text autocomplete |
autocompleteModel |
(empty) |
Dedicated model for autocomplete; empty = use the main model |
autocompleteDebounceMs |
200 |
Debounce delay before requesting a completion |
autocompleteContextLines |
15 |
Lines of prefix/suffix context sent per request |
Auto-verification
Runs diagnostics/typecheck/lint after file-modifying tool calls in Build/Debug mode and feeds the results back to the model for self-correction.
| Setting |
Default |
Description |
autoVerifyEnabled |
true |
Enable auto-verification |
autoVerifyTimeout |
60 |
Max seconds across all verification steps |
autoVerifyLintCommand |
(empty) |
Override the auto-detected lint command |
autoVerifyTestCommand |
(empty) |
Override the auto-detected test command (tests are opt-in — too slow to run by default) |
Web search
Used by the web_search tool. Auto-detected from whichever API key is present if searchProvider is left empty.
| Setting |
Default |
Description |
searchProvider |
(empty) |
Comma-separated providers to try: brave, serper, searxng, duckduckgo |
braveSearchApiKey |
(empty) |
Brave Search API key (free tier: 2000 queries/month) |
serperApiKey |
(empty) |
Serper (Google) API key |
searxngUrl |
(empty) |
Your SearXNG instance URL |
Project Structure
src/
├── agents/ # Sub-agent loop for multi-agent orchestration
├── api/ # LLM API client, prompt building, tool-call parsing
├── context/ # Token counting, mode management
├── conversation/ # Message flow orchestration, compaction
├── edit/ # Inline edits, autocomplete
├── indexing/ # Codebase indexer, retrieval service
├── observability/ # Request logging, session stats
├── provider/ # Model provider management
├── storage/ # Chat persistence
├── tools/ # Tool system
│ ├── executors/ # Individual tool implementations
│ ├── toolRegistry.ts
│ ├── policyEngine.ts
│ └── riskModel.ts
├── webview/ # HTML template, message routing
└── test/ # Unit tests + smoke test runner
media/ # Webview UI (main.js, main.css, icons)
Development
npm install
npm run compile # tsc -p ./
npm run lint # eslint src --ext ts
npm run test # full test suite (VS Code Electron host)
Press F5 in VS Code to launch an Extension Development Host for manual testing and debugging.
License
MIT
