An AI coding assistant with integrated multimodal vision capabilities
中文 | English
This is a personal fork of MiMoCode with a key enhancement: Multimodal Bridge — enabling the text-only mimo-v2.5-pro model to understand images by leveraging mimo-v2.5's vision capabilities.
mimo-v2.5-pro is a powerful code reasoning model, but it's text-only — it cannot process images directly. In real-world development, we often need to analyze UI screenshots, understand error messages from images, review architecture diagrams, or parse code snippets from screenshots. The Multimodal Bridge solves this by creating a seamless two-model pipeline.
User sends image + text
│
▼
mimo-v2.5 (Vision Model) ──→ Analyzes images, generates text description
│
▼
mimo-v2.5-pro (Reasoning Model) ──→ Receives text description, reasons about visual content
│
▼
Intelligent Response ──→ Code analysis, bug fixes, architecture guidance
- Image Detection — When you send an image with your prompt, the system detects image attachments in the message
- Vision Analysis — The image is sent to
mimo-v2.5(vision-capable) with a prompt asking for detailed description, focusing on code-related details - Text Bridge — The vision model's description replaces the image in the message as
[多模态视觉分析]prefixed text - Reasoning —
mimo-v2.5-proreceives the text description and reasons about the visual content with its full coding intelligence
The bridge is implemented as AI SDK middleware in the message pipeline. It activates automatically when all three conditions are met:
- Provider is
mimoorxiaomi - Model ID contains
v2.5-pro - Message content contains image parts
Core implementation in packages/opencode/src/provider/transform.ts:
export async function multimodalBridge(
msgs: ModelMessage[],
visionModel: LanguageModelV3,
): Promise<ModelMessage[]> {
// Extract all image parts from user messages
// ...
// Send all images to the vision model in one call
const { text } = await generateText({
model: visionModel,
messages: [{
role: "user",
content: [
...visionContent,
{ type: "text", text: "请详细描述以上所有图片的内容，保留对编程和代码相关的重要细节..." },
],
}],
})
// Replace image parts with the vision model's text description
// Each image part becomes: { type: "text", text: "[多模态视觉分析] " + description }
}Option 1: TUI Toggle (Recommended)
In the MiMoCode TUI, press Ctrl+p to open the command palette, find "多模态整合" (Multimodal Integration) and click to toggle. Or use the slash command:
/multimodal
Option 2: Config File
Add to .mimocode/mimocode.json:
{
"experimental": {
"multimodal_bridge": true
}
}Option 3: Environment Variable
export MIMOCODE_MULTIMODAL_BRIDGE=trueNote: The multimodal bridge is disabled by default. It only activates when using
mimo-v2.5-proas the main model — other models (Claude, GPT, etc.) already support images natively.
# One-line install
curl -fsSL https://mimo.xiaomi.com/install | bash
# Or install via npm
npm install -g @mimo-ai/cliThe first launch guides you through configuration automatically. Supported options:
- MiMo Auto (free for a limited time) — anonymous channel, zero configuration
- Xiaomi MiMo Platform — OAuth login
- Import from Claude Code — migrate existing authentication in one step
- Custom Provider — add any OpenAI-compatible API in the TUI
| Agent | Description |
|---|---|
| build | Default. Full tool permissions for development |
| plan | Read-only analysis mode for code exploration and solution design |
| compose | Orchestration mode for specs-driven development and skill-driven workflows |
Press Tab to switch between primary agents. Subagents are created by the system as needed.
Cross-session memory powered by SQLite FTS5 full-text search:
- Project memory (
MEMORY.md) — persistent project knowledge, rules, and architecture decisions - Session checkpoint (
checkpoint.md) — structured state snapshots maintained automatically by the checkpoint-writer subagent - Scratch notes (
notes.md) — temporary note area for agents - Task progress (
tasks/<id>/progress.md) — per-task logs
Memory is injected automatically when a session resumes, so the agent does not need to relearn project context.
- Automatic checkpoints — decides when to save session state based on the model context window
- Context reconstruction — when context approaches the limit, rebuilds it from the latest checkpoint, project memory, task progress, and retained recent messages so the agent can continue the current task
- Budgeted injection — uses a token budget to control how much checkpoint, memory, and notes content enters context, with importance ranking
A tree-shaped task system (T1, T1.1, T1.2, …) that integrates automatically with the checkpoint system, so task progress is preserved when sessions resume.
The primary agent can create subagents on demand. Subagents share the current session context and can work in parallel, with lifecycle tracking, cancellation, and background execution.
The /goal command sets a stopping condition for a session. When the agent tries to stop, an independent judge model evaluates the conversation to decide whether the condition is truly satisfied — preventing premature "optimistic stops" during autonomous work.
Compose mode provides a structured workflow for specs-driven development. It includes built-in skills for planning, execution, code review, TDD, debugging, verification, and merging — orchestrating the full lifecycle from spec to shipped code.
Real-time streaming voice input powered by TenVAD and MiMo ASR. Activate with /voice, then speak — audio is segmented by pauses and transcribed incrementally into the input. Available for MiMo logged-in users.
/dream— scans recent session traces, extracts persistent knowledge into project memory, and removes outdated entries/distill— discovers repeated manual workflows in recent work and packages high-confidence candidates into reusable skills, subagents, or commands
MiMoCode is configured via .mimocode/mimocode.json in the project directory (or ~/.config/mimocode/mimocode.json globally). Key options include:
- Provider and model selection
- Agent permissions and custom agents
- Checkpoint and memory behavior
- MCP server connections
- Keybindings and theme
- Multimodal bridge (
experimental.multimodal_bridge) — enable vision for text-only models
Max Mode (parallel best-of-N reasoning with judge selection) can be enabled via experimental.maxMode in the config.
bun install # Install dependencies
bun run dev # Run in development mode
bun turbo typecheck # Type checkThis project is a personal fork of MiMoCode, which is itself built as a fork of OpenCode. It keeps all core MiMoCode capabilities and adds the Multimodal Bridge feature.
Source code is licensed under the MIT License.
Use of MiMoCode is also subject to the Use Restrictions. Use of Xiaomi MiMo-hosted services is subject to the MiMo Terms of Service. Use of the MiMo name, logo, and trademarks is subject to the MiMo Trademark Policy.
- MiMoCode — The base project
- OpenCode — The original fork origin
- Xiaomi MiMo — For the powerful mimo-v2.5-pro and mimo-v2.5 models