Claude Sidecar

2026-05-15 · source · source ↗ · raw capture · wiki page

A parallel AI window that runs alongside Claude Code or Claude Cowork. Shares your conversation context, lets you talk to other models (Gemini, GPT, DeepSeek, Qwen, Grok, etc.) simultaneously, then folds results back into your main session.

Author: John Renaldi (GitHub: jrenaldi79)
Website: claudesidecar.ai
License: MIT
Stars: 13 | Forks: 7 | Open Issues: 10
Language: JavaScript (91.2%), HTML (7.8%), Shell (1.0%)
Latest Release: v0.5.2 (March 15, 2026)
Total Releases: 15 | Commits: 294

What It Does

Context Sharing — Automatically reads your Claude Code session history from ~/.claude/projects/[project]/[session].jsonl and passes it to the sidecar model. No copy-pasting needed.
Parallel Interaction — Main Claude session keeps working while you interact with the sidecar simultaneously.
Multi-Model Support — Connects to any model via OpenRouter or direct API keys. Ships with aliases for Gemini, GPT, Claude, DeepSeek, Qwen, Mistral, Grok, Kimi, GLM, MiniMax, and Seed.
Fold Mechanism — Click FOLD (or Cmd+Shift+F) to generate a structured summary that flows back into Claude Code context.
Two Modes — Interactive (Electron UI window) or Headless (--no-ui for autonomous background work).
Agent Modes — Chat (human-in-the-loop), Plan (read-only review), Build (full autonomy with write/bash access).
MCP Integration — Exposes tools (sidecar_start, sidecar_status, sidecar_read, etc.) as an MCP server, auto-registered for Claude Desktop and Cowork.
Session Persistence — List, read, resume, or chain sessions.
Adaptive Personality — Detects launch context (Claude Code vs. Cowork) and adjusts behavior accordingly.
Safety Features — Conflict detection, drift awareness, pre-flight validation.
Auto-Update — Checks npm registry daily; one-click update in UI.

Architecture

Sidecar is a harness built on OpenCode (opencode.ai), which provides the conversation runtime, tool execution, agent system, and web UI. Sidecar adds context sharing from Claude Code, the Electron shell, fold/summary workflow, session persistence, and multi-client support.

The Electron shell uses a BrowserView architecture: OpenCode web UI loads in a dedicated viewport, toolbar renders in the bottom 40px.

Installation

npm install -g claude-sidecar
sidecar setup  # graphical setup wizard for API keys, default model, routing

Commands

sidecar start --model <model> --prompt "<task>" — Launch a sidecar
sidecar list — Browse past sessions
sidecar resume <task_id> — Reopen a session
sidecar continue <task_id> --prompt "..." — New session building on previous
sidecar read <task_id> — Read session output
sidecar abort <task_id> — Stop a running session
sidecar setup — Configure API keys, models, routing
sidecar update — Update to latest version

Model Aliases

Alias	Model
`gemini`	Gemini 3.1 Flash Lite
`gemini-pro`	Gemini 3.1 Pro (1M context)
`gpt`	GPT-5.4
`codex`	GPT-5.3 Codex
`claude` / `sonnet`	Claude Sonnet 4.6
`opus`	Claude Opus 4.6
`deepseek`	DeepSeek v3.2
`qwen`	Qwen 3.5 397B
`grok`	Grok 4.1 Fast

MCP Tools

Tool	Description
`sidecar_start`	Spawn a sidecar (returns task ID immediately)
`sidecar_status`	Poll for completion
`sidecar_read`	Get results
`sidecar_list`	List past sessions
`sidecar_resume`	Reopen a session
`sidecar_continue`	New session building on previous
`sidecar_abort`	Stop a running session
`sidecar_setup`	Open setup wizard
`sidecar_guide`	Get usage instructions

Use Cases

Fact-check Claude's work
Get a second opinion on a feature
Deep-dive without polluting context
Parallel investigation (implement + review simultaneously)
Leverage model strengths (Gemini for 1M context, GPT for code gen, DeepSeek for cost-effective reasoning)