From bde5ab57436aab3c004cedf874fd6e139e9e9882 Mon Sep 17 00:00:00 2001 From: Claude Date: Sat, 3 Oct 2026 15:26:43 +0000 Subject: [PATCH] docs: publish under the Tromset account on ollama.com The GGUF tags of 0.2 beta are now online at ollama.com/Tromset/flycoder. The MLX tags still have to be pushed from an Apple Silicon Mac. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01CvbvR3WLYBpfyoeAE9sdqu --- Documentation.md | 18 +++++++++--------- brain/flybrain.mjs | 2 +- brain/router.mjs | 2 +- docs/ollama-model-page.md | 6 +++--- readme.md | 14 ++++++++------ tests/router.test.mjs | 2 +- 6 files changed, 23 insertions(+), 21 deletions(-) diff --git a/Documentation.md b/Documentation.md index 57175b1..24c9f92 100644 --- a/Documentation.md +++ b/Documentation.md @@ -47,16 +47,16 @@ ollama --version ### 3.1 From ollama.com -FlyCoder is published at [ollama.com/delairvictor9/flycoder](https://ollama.com/delairvictor9/flycoder). One command is enough: +FlyCoder is published at [ollama.com/Tromset/flycoder](https://ollama.com/Tromset/flycoder). One command is enough: ```sh -ollama run delairvictor9/flycoder # main variant (Gemma 4 12B, MLX) -ollama run delairvictor9/flycoder:fast # fast variant (Qwen3.5 4B, MLX) -ollama run delairvictor9/flycoder:0.2-beta-gguf # main, Intel Mac, Linux, Windows -ollama run delairvictor9/flycoder:0.2-beta-fast-gguf # fast, Intel Mac, Linux, Windows +ollama run Tromset/flycoder # main variant (Gemma 4 12B, MLX) +ollama run Tromset/flycoder:fast # fast variant (Qwen3.5 4B, MLX) +ollama run Tromset/flycoder:0.2-beta-gguf # main, Intel Mac, Linux, Windows +ollama run Tromset/flycoder:0.2-beta-fast-gguf # fast, Intel Mac, Linux, Windows ``` -Available tags: `latest` and `0.2-beta` (main), `fast` and `0.2-beta-fast` (fast), `0.2-beta-gguf` and `0.2-beta-fast-gguf` (portable). To call it simply `flycoder`: `ollama cp delairvictor9/flycoder flycoder`. +Available tags: `latest` and `0.2-beta` (main), `fast` and `0.2-beta-fast` (fast), `0.2-beta-gguf` and `0.2-beta-fast-gguf` (portable). For now only the GGUF tags (`0.2-beta-gguf`, `0.2-beta-fast-gguf`, `0.2-beta-lite-gguf`, `router-gguf`) are online; the MLX tags arrive once they are pushed from an Apple Silicon Mac. Until then, Mac users install with `install.sh`. To call it simply `flycoder`: `ollama cp Tromset/flycoder flycoder`. ### 3.2 One-command installer @@ -205,7 +205,7 @@ ollama launch codex --model flycoder # Codex ollama launch opencode --model flycoder # OpenCode ``` -If you got FlyCoder from ollama.com without `install.sh`, use the published name: `ollama launch claude --model delairvictor9/flycoder`. +If you got FlyCoder from ollama.com without `install.sh`, use the published name: `ollama launch claude --model Tromset/flycoder`. `ollama launch --help` lists the other integrations, including VS Code, Copilot CLI, Cline, Qwen Code, Pi and Droid. Agents send long instructions and many files: Ollama recommends at least 64,000 tokens of context for them (section 6.2). @@ -215,7 +215,7 @@ FlyBrain is a small, dependency-free Node 22 server that sits in front of Ollama ```sh node ~/.flycoder/brain/flybrain.mjs # copied by install.sh; or npm run brain in this repository -node brain/flybrain.mjs --prefix delairvictor9/ # with the models published on ollama.com +node brain/flybrain.mjs --prefix Tromset/ # with the models published on ollama.com ``` Options: `--port` (11435), `--ollama` (http://127.0.0.1:11434), `--prefix`, `--max-expert auto|full|fast`. With `auto`, the default, a Mac with less than 16 GB caps `flycoder` at the 4B. @@ -336,7 +336,7 @@ On this server, the main variant solves almost three times as many problems, but ## 8. Publishing a version on ollama.com -0.2 beta is published under the `delairvictor9` account. To publish a new version, follow these steps. The script refuses to push non-MLX builds under the main tags: build them on an Apple Silicon Mac. +The GGUF tags of 0.2 beta are published under the `Tromset` account; the MLX tags must be pushed from an Apple Silicon Mac, because Ollama cannot build MLX models on Linux. To publish a new version, follow these steps. The script refuses to push non-MLX builds under the main tags: build them on an Apple Silicon Mac. 1. Create an account on [ollama.com](https://ollama.com/signup). The username becomes the model's prefix (`/flycoder`). 2. Link the machine to the account: `ollama signin`, then confirm in the browser. Authorized machines are managed in the account settings, under Ollama keys; remove the ones you no longer use. diff --git a/brain/flybrain.mjs b/brain/flybrain.mjs index 9e88ce7..c8d376d 100644 --- a/brain/flybrain.mjs +++ b/brain/flybrain.mjs @@ -1,7 +1,7 @@ #!/usr/bin/env node // FlyBrain: an Ollama-compatible server that answers to `flycoder` and `flycoder:fast`, // routes each request to one expert model and keeps a single expert in memory. -// Usage: node brain/flybrain.mjs [--port 11435] [--ollama http://127.0.0.1:11434] [--prefix delairvictor9/] [--max-expert auto|full|fast] +// Usage: node brain/flybrain.mjs [--port 11435] [--ollama http://127.0.0.1:11434] [--prefix Tromset/] [--max-expert auto|full|fast] import http from 'node:http'; import os from 'node:os'; import fs from 'node:fs'; diff --git a/brain/router.mjs b/brain/router.mjs index 870c691..3f9f784 100644 --- a/brain/router.mjs +++ b/brain/router.mjs @@ -37,7 +37,7 @@ export function ruleRoute(d) { } // Experts behind each name FlyBrain answers to. `prefix` lets the published -// models be used directly, e.g. prefix 'delairvictor9/' for delairvictor9/flycoder:0.2-beta. +// models be used directly, e.g. prefix 'Tromset/' for Tromset/flycoder:0.2-beta. export function profiles(prefix = '', { maxExpert = 'full' } = {}) { const tag = t => `${prefix}flycoder:${t}`; const normal = { simple: tag('0.2-beta-fast'), hard: tag(maxExpert === 'fast' ? '0.2-beta-fast' : '0.2-beta'), router: tag('router'), think: true }; diff --git a/docs/ollama-model-page.md b/docs/ollama-model-page.md index 53ed2e4..8266c58 100644 --- a/docs/ollama-model-page.md +++ b/docs/ollama-model-page.md @@ -3,8 +3,8 @@ A local coding model tuned for MacBook and Mac mini (Apple Silicon), packaged for Ollama. ```sh -ollama run delairvictor9/flycoder # Gemma 4 12B base, Macs with 16 GB or more -ollama run delairvictor9/flycoder:fast # Qwen3.5 4B base, Macs with 8 GB +ollama run Tromset/flycoder # Gemma 4 12B base, Macs with 16 GB or more +ollama run Tromset/flycoder:fast # Qwen3.5 4B base, Macs with 8 GB ``` ## Tags @@ -32,7 +32,7 @@ MLX tags need Ollama 0.31 or later on Apple Silicon. FlyBrain is a small Node server from the GitHub repository that sits in front of Ollama. It answers to `flycoder` and `flycoder:fast`, sends simple questions to a smaller expert, sends hard requests (and every agent request with tools) to the strong one, and keeps a single expert in memory. Measured on a 16 GB Linux server (GGUF, CPU): a simple request to `flycoder` uses 6.1 GiB instead of 10.5. Hard requests still use the strong model, so their quality is unchanged. ```sh -node brain/flybrain.mjs --prefix delairvictor9/ +node brain/flybrain.mjs --prefix Tromset/ OLLAMA_HOST=127.0.0.1:11435 ollama run flycoder ``` diff --git a/readme.md b/readme.md index 9d92823..0a0ff6e 100644 --- a/readme.md +++ b/readme.md @@ -6,13 +6,15 @@ Full documentation (installation, usage, API, settings, benchmark, publishing, t ## One-command install -You need [Ollama](https://ollama.com/download) 0.31 or later, running. FlyCoder is published at [ollama.com/delairvictor9/flycoder](https://ollama.com/delairvictor9/flycoder): +You need [Ollama](https://ollama.com/download) 0.31 or later, running. FlyCoder is published at [ollama.com/Tromset/flycoder](https://ollama.com/Tromset/flycoder): ```sh -ollama run delairvictor9/flycoder # Gemma 4 12B, Macs with 16 GB or more -ollama run delairvictor9/flycoder:fast # Qwen3.5 4B, 8 GB Macs +ollama run Tromset/flycoder # Gemma 4 12B, Macs with 16 GB or more +ollama run Tromset/flycoder:fast # Qwen3.5 4B, 8 GB Macs ``` +For now only the portable GGUF tags are online (`ollama run Tromset/flycoder:0.2-beta-gguf`); the Mac (MLX) tags above come next. Until then, Macs install with `install.sh` below. + Or, to install it under the short name `flycoder`, with the variant picked from your Mac's memory: ```sh @@ -63,7 +65,7 @@ ANTHROPIC_BASE_URL=http://127.0.0.1:11435 ANTHROPIC_AUTH_TOKEN=ollama ANTHROPIC_ # Codex, OpenCode, Cline…: OpenAI-compatible URL http://127.0.0.1:11435/v1, model flycoder ``` -With the models published on ollama.com rather than installed by `install.sh`: `ollama pull delairvictor9/flycoder:router` (and `:0.2-beta-lite`, `:0.2-beta-fast`, `:0.2-beta`), then `node brain/flybrain.mjs --prefix delairvictor9/`. +With the models published on ollama.com rather than installed by `install.sh`: `ollama pull Tromset/flycoder:router` (and `:0.2-beta-lite`, `:0.2-beta-fast`, `:0.2-beta`), then `node brain/flybrain.mjs --prefix Tromset/`. ## What changed since 0.1 @@ -116,7 +118,7 @@ ollama run flycoder --think=false # immediate answer, for simple question ollama run flycoder:fast # 4B variant if you installed it ``` -In a conversation, `/set nothink` turns thinking off and `/set parameter num_ctx 65536` enlarges the context (about 0.5 GB of extra memory for the main variant). Ollama recommends at least 64,000 tokens for coding agents; FlyCoder also works with `ollama launch` (Claude Code, Codex, OpenCode): `ollama launch claude --model flycoder` if you installed it with `install.sh`, or `ollama launch claude --model delairvictor9/flycoder` from ollama.com. `ollama launch` talks to Ollama directly; to go through FlyBrain, see the FlyBrain section. +In a conversation, `/set nothink` turns thinking off and `/set parameter num_ctx 65536` enlarges the context (about 0.5 GB of extra memory for the main variant). Ollama recommends at least 64,000 tokens for coding agents; FlyCoder also works with `ollama launch` (Claude Code, Codex, OpenCode): `ollama launch claude --model flycoder` if you installed it with `install.sh`, or `ollama launch claude --model Tromset/flycoder` from ollama.com. `ollama launch` talks to Ollama directly; to go through FlyBrain, see the FlyBrain section. From an application, the Ollama API is all you need: @@ -134,7 +136,7 @@ Done once, from an Apple Silicon Mac: 4. Publish: `sh scripts/publish.sh ` (add `--gguf` to also publish the Intel Mac, Linux and Windows builds). 5. Paste the text of [docs/ollama-model-page.md](docs/ollama-model-page.md) into the model page description. -After that, anyone can run `ollama run /flycoder` or `ollama run /flycoder:fast`. 0.2 beta is published under `delairvictor9`. +After that, anyone can run `ollama run /flycoder` or `ollama run /flycoder:fast`. The GGUF tags of 0.2 beta are published under `Tromset`; the MLX tags (`latest`, `fast`) still have to be pushed from an Apple Silicon Mac. ## Repository contents diff --git a/tests/router.test.mjs b/tests/router.test.mjs index c743a7d..98ab30a 100644 --- a/tests/router.test.mjs +++ b/tests/router.test.mjs @@ -56,7 +56,7 @@ test('route: a conversation keeps its expert and can only move up', async () => }); test('profiles: prefix for published models and the 8 GB cap', () => { - assert.equal(profiles('delairvictor9/').flycoder.hard, 'delairvictor9/flycoder:0.2-beta'); + assert.equal(profiles('Tromset/').flycoder.hard, 'Tromset/flycoder:0.2-beta'); assert.equal(profiles('', { maxExpert: 'fast' }).flycoder.hard, 'flycoder:0.2-beta-fast'); assert.equal(defaultMaxExpert(8 * 2 ** 30), 'fast'); assert.equal(defaultMaxExpert(16 * 2 ** 30), 'full');