|
| 1 | +# 📋 Migration Note: mlx-server → SwiftLM |
| 2 | + |
| 3 | +> **Date**: 2026-03-30 |
| 4 | +> **Affects**: All Aegis-AI components that reference the MLX inference sidecar |
| 5 | +
|
| 6 | +--- |
| 7 | + |
| 8 | +## What Changed |
| 9 | + |
| 10 | +The MLX inference engine previously known as **`mlx-server`** has been renamed to **`SwiftLM`**. |
| 11 | + |
| 12 | +This is a **rename only** — the API, ports, endpoints, and behaviour are 100% identical. |
| 13 | + |
| 14 | +--- |
| 15 | + |
| 16 | +## Required Updates in Aegis-AI |
| 17 | + |
| 18 | +### 1. Binary path / process spawn |
| 19 | + |
| 20 | +| Before | After | |
| 21 | +|---|---| |
| 22 | +| `/path/to/mlx-server` | `/path/to/SwiftLM` | |
| 23 | +| `.build/release/mlx-server` | `.build/release/SwiftLM` | |
| 24 | + |
| 25 | +Update any hardcoded binary paths in: |
| 26 | +- Engine registry / scheduler config |
| 27 | +- IPC process spawn calls (`spawn('mlx-server', ...)`) |
| 28 | +- systemd / launchd service definitions (if any) |
| 29 | + |
| 30 | +### 2. GitHub release URL |
| 31 | + |
| 32 | +| Before | After | |
| 33 | +|---|---| |
| 34 | +| `https://github.com/SharpAI/mlx-server/releases` | `https://github.com/SharpAI/SwiftLM/releases` | |
| 35 | +| `mlx-server-*-macos-arm64.tar.gz` | `SwiftLM-*-macos-arm64.tar.gz` | |
| 36 | + |
| 37 | +### 3. Wisdom cache directory |
| 38 | + |
| 39 | +| Before | After | |
| 40 | +|---|---| |
| 41 | +| `~/.mlx-server/wisdom/` | `~/.swiftlm/wisdom/` | |
| 42 | + |
| 43 | +If the user has existing calibration wisdom at `~/.mlx-server/wisdom/`, migrate it: |
| 44 | + |
| 45 | +```bash |
| 46 | +mkdir -p ~/.swiftlm |
| 47 | +cp -r ~/.mlx-server/wisdom ~/.swiftlm/wisdom |
| 48 | +``` |
| 49 | + |
| 50 | +### 4. Log prefix in console output |
| 51 | + |
| 52 | +| Before | After | |
| 53 | +|---|---| |
| 54 | +| `[mlx-server]` | `[SwiftLM]` | |
| 55 | + |
| 56 | +Update any log parsers or regex filters watching for `[mlx-server]` in stdout. |
| 57 | + |
| 58 | +### 5. UI display name |
| 59 | + |
| 60 | +Any UI label showing `"mlx-server"` or `"MLX Server"` should be updated to `"SwiftLM"`. |
| 61 | + |
| 62 | +--- |
| 63 | + |
| 64 | +## What Did NOT Change |
| 65 | + |
| 66 | +| Item | Value | |
| 67 | +|---|---| |
| 68 | +| Default port | `5413` (unchanged) | |
| 69 | +| API endpoints | `/v1/chat/completions`, `/health`, `/v1/models` (unchanged) | |
| 70 | +| CLI flags | All flags identical (`--model`, `--stream-experts`, `--turbo-kv`, etc.) | |
| 71 | +| Ready event JSON | `{"event":"ready", ...}` (unchanged) | |
| 72 | +| OpenAI compatibility | Full (unchanged) | |
| 73 | +| Model format (MLX/HF) | Unchanged | |
| 74 | + |
| 75 | +--- |
| 76 | + |
| 77 | +## No Action Required If… |
| 78 | + |
| 79 | +- You are using the **OpenAI SDK** pointed at `http://127.0.0.1:5413/v1` — zero changes needed |
| 80 | +- You are using **`AEGIS_INTEGRATION.md`** as your integration reference — it has already been updated |
| 81 | + |
| 82 | +--- |
| 83 | + |
| 84 | +## Git History Note |
| 85 | + |
| 86 | +All historical commits previously attributed to `Aegis-AI <aegis@sharpai.com>` have been rewritten to `Simba Zhang <solderzzc@gmail.com>`. If you have a local clone of the old `mlx-server` repo, you will need to re-clone `SwiftLM` or run: |
| 87 | + |
| 88 | +```bash |
| 89 | +git remote set-url origin https://github.com/SharpAI/SwiftLM.git |
| 90 | +git fetch --all |
| 91 | +git reset --hard origin/main |
| 92 | +``` |
0 commit comments