model-mapper 1.3

Added

• vision_mappings per upstream for routing image requests to image-capable models.
• Xiaomi MiMo vision model support via mimo-v2-omni on the Anthropic-compatible endpoint.
• Web UI fields for configuring text and image models separately per tier.
• Regression tests for image request detection, vision routing, and missing vision-model errors.

Changed

• Requests containing Anthropic image blocks now use vision_mappings instead of text mappings.
• Sample config and README now document text versus image model mappings.

Fixed

• Image requests no longer get sent to text-only upstream models such as mimo-v2.5-pro.
• Upstreams without vision mappings now return a clear configuration error for image requests instead of forwarding to an unsupported model.

model-mapper 1.2

Added

• Explicit model_routes config for choosing which upstream handles each client-facing model.
• Web UI routing table preserves each upstream's full model catalog while allowing per-tier mixed routing.
• Regression test for mixed routing when multiple upstreams define the same client model names.
• MODEL_MAPPER_CONFIG environment variable for selecting a config file path.

Changed

• Request dispatch resolves upstream selection from model_routes first, then falls back to default_upstream.
• Documentation and sample config now describe model_routes + upstreams[].mappings[].

Fixed

• New processes started from the project directory read the local config.json.
• Web UI add-upstream flow works after fixing a JavaScript parse error.
• Mixed upstream routing no longer collapses all model tiers onto the global default upstream.

v1.1 — Multi-upstream routing

Highlights

• 多上游路由：按模型名自动分发到不同上游，支持 DeepSeek、MiMo 等多厂商
• 三档路由表：Opus / Sonnet / Haiku 固定三列，切换上游只需改下拉框
• 上游卡片：每个上游独立配置 URL/Token/认证方式/协议，内置三档模型名映射
• 自动协议转换：上游协议（Anthropic/OpenAI）与客户端不匹配时自动转换
• 优雅关闭：SIGTERM 后等待当前请求完成
• 健康检查：/healthz 端点
• 连接超时：Dial + TLS 握手超时防止挂起

Upgrading from 1.0

旧 config.json 启动时自动迁移。无需手动操作。

v1.0.0

Initial release

Features

• Passthrough mode for Anthropic-compatible upstreams (DeepSeek, etc.)
• Protocol conversion mode (Anthropic → OpenAI) for OpenAI-compatible upstreams
• Multi-target model mapping (e.g. claude-opus-4-6 → deepseek-v4-pro, claude-sonnet-4-6 → deepseek-v4-flash)
• Web UI for configuration
• Systemd + nginx deployment support

Binaries

• model-mapper-darwin-arm64 — macOS Apple Silicon
• model-mapper-linux-amd64 — Linux x86_64