OMP profiles
What a profile actually is, which fields OOMPF reads, and where the runtime authority stays.
Share a working OMP profile as one link.
Profiles get passed around as screenshots and half a YAML file. OOMPF indexes them from public GitHub Gists, shows the facts —models, providers, advisors, hooks, prerequisites— and hands back a command that installs the profile natively.
oompf publish daily-driver ✓ gist gist.github.com/alice/cfcdad13… ✓ your link oompf.run/p/prof_00e08f39… — share itoompf add https://oompf.run/p/prof_00e08f39… ✓ installed ~/.omp/profiles/alice-daily-driveromp --profile alice-daily-driver
npm install -g @grosspoetrysystems/oompfoompf publish <profile>Two commands and your profile has a permanent link with its models, providers, and prerequisites rendered — yours to paste anywhere. The YAML stays on GitHub as a public Gist; OOMPF indexes the facts. How publishing works
Recently shared
sub-codex@thekidnamedkdgeneral
One provider, differentiated by task — the max-reasoning tier for planning, the standard tier for the everyday, a cheaper tier for chores. All OpenAI Codex, so it runs in a single-provider harness (Codex) as well as omp/pi. Reach for it when you are all-in on one provider but still want per-task model selection.
modelsopenai-codex/gpt-5.6-luna · openai-codex/gpt-5.6-luna:max · openai-codex/gpt-5.5
providersopenai-codex
installoompf add https://oompf.run/p/prof_8d14c0854bfdd11cf75a5f0226b6441a
sub-claude@thekidnamedkdgeneralOne provider, differentiated by task — Opus for planning, Sonnet for the everyday, Haiku for cheap chores. All Anthropic, so it runs anywhere Claude runs (Claude Code included), not only multi-provider harnesses. Reach for it when you live in one provider but still want the right model per task.
modelsanthropic/claude-sonnet-4.6 · anthropic/claude-opus-4-8 · anthropic/claude-haiku-4.6
providersanthropic
installoompf add https://oompf.run/p/prof_db9da061f373a7fc8612ddae8c8cd423
research@thekidnamedkdresearchOptimized for investigation — frontier reasoning by default, Opus for planning the approach, and generous thinking budgets. Reach for it when the goal is understanding a problem deeply rather than shipping code fast.
modelsopenai/gpt-5.6 · anthropic/claude-opus-4-8 · anthropic/claude-sonnet-4.6 +1
providersopenai · anthropic
installoompf add https://oompf.run/p/prof_3858e8c9fe61d2b5bcbf207be98db8b8
budget@thekidnamedkdbudgetThe cheapest way to keep working — DeepSeek Flash for the bulk, a local model for chores, Kimi for planning, GLM for review. Reach for it on high-volume, low-stakes work where throughput per dollar matters more than peak quality.
modelsdeepseek/deepseek-v4-flash · ollama/qwen3.6 · moonshotai/kimi-k2.6 +1
providersdeepseek · ollama · moonshotai · z-ai
installoompf add https://oompf.run/p/prof_d2c4ffc29a30f50f68cedfd1bb1139c8
max-quality@thekidnamedkdgeneralNo compromises — Opus and GPT-5.6 at high reasoning for everything, an independent frontier reviewer, and a dedicated security pass. Reach for it on high-stakes work where a wrong answer costs far more than the tokens.
modelsanthropic/claude-opus-4-8 · openai/gpt-5.6 · google/gemini-3.5-pro +1
providersanthropic · openai · google
installoompf add https://oompf.run/p/prof_446297f3a63be7851ffb8aaecebd8f1b
Start here
OOMPF indexes OMP's own profile format and adds one metadata convention inside it. OMP remains the authority for how a profile behaves once it is installed. What OOMPF is, and is not
Publishing
Turn a local profile into a public Gist, scan it for secrets, and register its metadata.
Installing
Install someone else's profile by its OOMPF URL, and what happens when the name collides.