turbollm 1.7.7 → 1.8.1

package/README.md

@@ -12,6 +12,7 @@
 </p>
 
 <p align="center">
+  <a href="https://turbollm.dev"><img src="https://img.shields.io/badge/turbollm.dev-docs%20·%20models%20·%20guides-e2552e" alt="turbollm.dev" /></a>
   <a href="https://www.npmjs.com/package/turbollm"><img src="https://img.shields.io/npm/v/turbollm.svg?color=e2552e" alt="npm version" /></a>
   <a href="https://www.npmjs.com/package/turbollm"><img src="https://img.shields.io/npm/dm/turbollm.svg?color=e2552e" alt="npm downloads" /></a>
   <img src="https://img.shields.io/badge/node-%E2%89%A522-3c873a.svg" alt="node >= 22" />
@@ -22,6 +23,11 @@
   <a href="https://discord.gg/v6kRbV7nC"><img src="https://img.shields.io/badge/Discord-join%20chat-5865F2?logo=discord&logoColor=white" alt="Discord" /></a>
 </p>
 
+<p align="center">
+  <sub><strong>Source-available</strong> under <a href="#license">FSL-1.1</a> — free for personal and
+  internal business use; every release converts to <strong>Apache-2.0</strong> two years after it ships.</sub>
+</p>
+
 <!-- Brand: shipped app icon web/public/brand/turbollm-icon-512.jpeg · high-res masters web/brand-assets/ (unshipped) · in-app mark web/src/components/Logo.tsx · favicon web/public/favicon.svg -->
 
 ```bash
@@ -31,6 +37,12 @@ npx turbollm
 That one command starts a local daemon, opens a browser UI, and serves your models over an
 API any tool can talk to. TurboLLM is the **performance & bleeding-edge layer for local
 LLMs** — built for people who today hand-compile forks and hunt forums for the right flags.
+The speed is measured, not promised: **74.7 vs 61 t/s on the *same* official llama.cpp as
+LM Studio — and 2.2× on forks it can't load** ([the numbers](#speed-turbollm-vs-lm-studio)).
+
+<p align="center">
+  <strong>📖 Full docs · what runs on your GPU · model picks → <a href="https://turbollm.dev">turbollm.dev</a></strong>
+</p>
 
 <p align="center">
   <img src="https://raw.githubusercontent.com/mohitsoni48/TurboLLM/main/assets/how-it-works.svg?v=2" width="860" alt="How TurboLLM works: clients -> one lightweight daemon -> any engine on your GPU" />
@@ -41,6 +53,7 @@ LLMs** — built for people who today hand-compile forks and hunt forums for the
 ## Contents
 
 - [Why TurboLLM](#why-turbollm)
+- [Who this is for — and who it isn't](#who-this-is-for--and-who-it-isnt)
 - [Speed: TurboLLM vs LM Studio](#speed-turbollm-vs-lm-studio)
 - [Features](#features)
 - [Quick start](#quick-start)
@@ -79,7 +92,7 @@ TurboLLM does the opposite:
   a **VRAM-fit verdict before you load** — no more flag guessing.
 - **📊 Real tokens/sec, never faked.** Speed in the model list is *measured on your machine*
   from actual generation — live while you chat, and remembered per model.
-- **🪶 Lightweight.** A ~2 MB npm package on Node — **no Electron, no bundled Chromium, no
+- **🪶 Lightweight.** A ~7 MB npm package on Node — **no Electron, no bundled Chromium, no
   Python**. It downloads only the engine your GPU actually needs (Vulkan ≈ 38 MB).
 - **🔌 Drop-in APIs.** OpenAI **and** Anthropic-compatible — so Claude Code and every existing
   tool work unchanged.
@@ -90,6 +103,27 @@ TurboLLM does the opposite:
 
 ---
 
+## Who this is for — and who it isn't
+
+**It's for you if:**
+
+- you're tired of hand-tuning `-ngl` / KV / offload flags for every model and never trusting
+  the speed number;
+- you want community forks (low-bit KV cache, NextN speculative decoding, new quant formats)
+  on day 0 — without compiling anything;
+- you're pointing agents or **Claude Code** at a local model and want that to be one command;
+- you already have folders of GGUFs from LM Studio / Ollama and refuse to re-download them.
+
+**It's probably not for you if:**
+
+- you want a zero-terminal, one-click desktop installer — TurboLLM currently needs
+  [Node ≥22](#requirements) and one terminal command (a packaged installer is planned);
+- you're happy with a single blessed runtime and stock settings — LM Studio and Ollama
+  already serve that well;
+- you need training or fine-tuning — TurboLLM is inference only.
+
+---
+
 ## Speed: TurboLLM vs LM Studio
 
 Same GPU (RTX 5070 Ti 16 GB), same model, same 200K context — measured generation speed.
@@ -134,8 +168,9 @@ the detail:
 <br/>
 
 - **Use the folders you already have.** Point TurboLLM at any directory of GGUFs — your
-  existing LM Studio / Ollama / manual downloads — **no re-downloading.** It parses GGUF
-  metadata (arch, params, quant, context, vision) for every file.
+  existing LM Studio folders or manual downloads — **no re-downloading.** It parses GGUF
+  metadata (arch, params, quant, context, vision) for every file. (Models pulled by
+  `ollama pull` live in Ollama's own blob store, which TurboLLM doesn't read.)
 - **Browse & download from Hugging Face**, in-app: a live, sortable list (trending / downloads
   / likes / recently updated / newest) alongside a permanent detail pane — pick a quant, read
   the **rendered model card**, and download with **resume + SHA-256 verification**. Each model
@@ -183,6 +218,9 @@ the detail:
   overwrites another engine's tuning for the same model.
 - **Configurable VRAM headroom** (Settings → Models & loading → Advanced, 300 MB–2 GB, default
   1 GB) — tell auto-tune how much VRAM to keep free for other GPU workloads instead of a fixed margin.
+  Drag it to 0 to opt into an experimental **MoE "VRAM-spill" search** — auto-tune keeps pushing more
+  experts onto the GPU past the safe margin as long as both generation and prompt-processing speed
+  keep improving.
 
 </details>
 
@@ -236,6 +274,35 @@ the detail:
   in chat (name, preferences, hardware) using your own loaded model, and carries them into future
   new conversations. Nothing leaves your device; the full fact list is reviewable and deletable
   from Settings → Memory, and turning the toggle off stops new chats from seeing it immediately.
+- **Thinking-budget control** — a graduated slider, not just on/off: cap reasoning to a specific
+  token count, disable it entirely, or leave it unlimited. Works in Chat and Code alike.
+
+</details>
+
+<details>
+<summary><strong>🧑‍💻 Code — a local coding agent, in a real project directory</strong></summary>
+<em>(experimental, off by default — Settings → Experimental)</em>
+
+<br/>
+
+Workspace → Code hands a task to an agent running on the same model you already have loaded —
+point it at a repo folder, describe what you want, and it reads, edits, and runs commands to get
+there. Entirely local; nothing leaves your machine.
+
+- **Real repo access** — plans and edits end-to-end (or asks before mutating, or plan-only,
+  depending on mode), with an optional isolated git worktree so your actual checkout stays
+  untouched, and a real diff summary of what changed.
+- **Persistent sessions** — archive/filter past runs, revert to any earlier message (with
+  optional real file-edit reversal), attach files as context, and a "Coding activity" dashboard
+  (sessions, tasks shipped, files touched, diff shipped, streaks) built from real history, not
+  mock data.
+- **Real LSP integration** for TypeScript/JavaScript and Python — detects the language, installs
+  the language server if needed, and uses it for edits.
+- **The same tools Chat gets from Customize** — any connected MCP server, plus the sandboxed
+  `run_code` tool — are available to Code too, alongside honest skill invocation and
+  `AGENTS.md`/`agents.md` support.
+- **Independent access control** — gated behind its own API key on non-host devices, separate
+  from Chat's gate, and hidden entirely until you opt in from Settings → Experimental.
 
 </details>
 
@@ -316,7 +383,7 @@ connected); **Remove** undoes it.
 
 <br/>
 
-- A **~2 MB npm package** on Node — no Electron, no bundled Chromium, no Python.
+- A **~7 MB npm package** on Node — no Electron, no bundled Chromium, no Python.
 - **Offline-first** — no account, no backend, no internet, no telemetry.
 - **Windows · macOS · Linux**, with a CPU fallback when there's no GPU.
 
@@ -472,7 +539,22 @@ Use `--config <file>` to point at an alternate config (its directory becomes the
 
 ## Requirements
 
-- **Node.js 22.13.0 or newer** — enforced at startup with a clear message. <https://nodejs.org>
+- **Node.js 22.13.0 or newer** — enforced at startup with a clear message.
+
+  <details>
+  <summary><strong>Don't have Node?</strong> One command installs it.</summary>
+
+  <br/>
+
+  - **Windows:** `winget install OpenJS.NodeJS.LTS` (or download from <https://nodejs.org>)
+  - **macOS:** `brew install node` (or download from <https://nodejs.org>)
+  - **Linux:** use your distro's package manager or <https://nodejs.org> — make sure it's
+    v22+ (`node --version`)
+
+  Then open a **new terminal** (Windows needs one for `PATH` to refresh) and run `npx turbollm`.
+
+  </details>
+
 - **Windows, macOS, or Linux.**
 - A GPU is recommended but **not required** — a CPU build is provisioned as a fallback.
 - On Windows, the first time the auto-downloaded `llama-server` runs, SmartScreen/Defender may
@@ -505,7 +587,7 @@ Marks reflect mid-2026; verify the moving rows against each tool's current docs.
 | Speculative decoding (draft / MTP) | ✅ | ✅ | ◐ env flag | ❌ |
 | Web UI from any LAN device | ✅ | ❌ | ❌ | ✅ |
 | **Lightweight** (no Electron / no Python) | ✅ npm | ❌ Electron | ✅ Go | ❌ Python |
-| Offline-first · **no telemetry** | ✅ | ◐ analytics on by default | ✅ | ✅ |
+| Offline-first · **no telemetry** | ✅ verifiable (source-available) | ◐ closed app — not verifiable | ✅ | ✅ |
 
 LM Studio and Ollama both added Anthropic `/v1/messages` endpoints in 2026, so the API rows are
 now parity — Claude Code works against any of them. TurboLLM's durable edges are **any engine
@@ -553,14 +635,33 @@ shadcn/ui frontend. One TypeScript codebase, shipped as an npm package.
 ## Community
 
 Questions, ideas, and show-and-tell — join the [Discord](https://discord.gg/v6kRbV7nC).
+Want to contribute? See [CONTRIBUTING.md](https://github.com/mohitsoni48/TurboLLM/blob/main/CONTRIBUTING.md).
+
+If TurboLLM saved you flag-hunting time, a ⭐ on the repo helps others find it.
 
 ---
 
 ## License
 
 Source-available under the **Functional Source License 1.1 (Apache-2.0 future grant)** — SPDX
-**`FSL-1.1-ALv2`**. Free for personal use, internal business use, education, and research; the
-only restriction is shipping a competing product. Each release converts to Apache-2.0 two
-years after it's published. Full text: [LICENSE.md](https://github.com/mohitsoni48/TurboLLM/blob/main/turbollm/LICENSE.md).
+**`FSL-1.1-ALv2`**. Full text: [LICENSE.md](https://github.com/mohitsoni48/TurboLLM/blob/main/turbollm/LICENSE.md).
+
+**License FAQ** (plain-language summary — the license text is what's binding):
+
+- **Can I use it for free?** Yes. Personal use, internal business use, education, research,
+  and self-hosting — including commercially, inside your company — are all free. The license
+  permits *any* use except the one below.
+- **What's the one restriction?** You can't take TurboLLM and ship it (or its functionality)
+  as a **competing commercial product or service**. That's the entire boundary.
+- **Can I fork it or redistribute it?** Yes — modify, fork, and redistribute freely for any
+  permitted purpose; keep the license text and copyright notices with it.
+- **When does it become fully open source?** The license includes an **irrevocable** grant:
+  each release converts to **Apache-2.0 exactly two years** after that release is published.
+  The first public release shipped June 2026, so it converts in June 2028 — and every release
+  after it follows on its own two-year clock.
+- **Why not MIT/Apache from day one?** TurboLLM is built by a solo maintainer; FSL prevents a
+  large vendor from re-skinning it as their own product while it's young, and the future grant
+  guarantees the full open-source outcome anyway. (Sentry created the FSL; GitButler, PowerSync,
+  and Codecov ship under it too.)
 
 <p align="center"><sub>Built for people who refuse to wait for the mainstream to bless the fast path. ⚡</sub></p>

package/dist/{chunk-TPYU5ZAX.js → chunk-U4OPDHSX.js}

@@ -7,11 +7,17 @@ import { join } from "path";
 // src/engines/compat.ts
 function engineAcceptsFormat(engineKind, format) {
   if (engineKind === "mlx") return format === "mlx";
+  if (engineKind === "rapid-mlx") return format === "mlx";
   if (engineKind === "vllm") return format === "mlx";
   return format === "gguf";
 }
+function engineRejectsAudioModel(engineKind) {
+  return engineKind === "rapid-mlx";
+}
 var ENGINE_MODEL_ALIAS = "default_model";
+var RAPID_MLX_MODEL_ALIAS = "default";
 function engineModelAlias(engineKind) {
+  if (engineKind === "rapid-mlx") return RAPID_MLX_MODEL_ALIAS;
   return engineKind === "mlx" || engineKind === "vllm" || engineKind === "sglang" ? ENGINE_MODEL_ALIAS : null;
 }
 var VENDOR_DISPLAY = {
@@ -86,7 +92,7 @@ Write ONE reusable skill as raw JSON \u2014 no markdown fences.` }
         ],
         temperature: 0.3,
         max_tokens: 800,
-        reasoning_budget: 0,
+        thinking_budget_tokens: 0,
         chat_template_kwargs: { enable_thinking: false },
         response_format: { type: "json_object" },
         stream: false
@@ -168,6 +174,7 @@ ${f.content}`);
 
 export {
   engineAcceptsFormat,
+  engineRejectsAudioModel,
   engineModelAlias,
   evaluateVariant,
   distillFromConversation,