BLOSSOM ECOSYSTEM · OPEN SOURCE

One prompt.The right model.

Keep routine AI work local. Route harder tasks to the cloud only when they deserve the cost.

$ bloom auto --dry-run "design a migration plan"

ROUTE DECISIONstrongcomplexity keyword

01 · ROUTES

A small router, not another AI platform.

Blossom Router sits between your prompt and the models you already use. The rules are visible, local-first, and cheap to run.

Local

Short questions and private routine work

Ollama

Cheap

Summaries, translation, formatting, batch text

Cloud · low cost

Normal

Everyday coding and general cloud work

Cloud · balanced

Strong

Architecture, security, migration, root cause

Cloud · premium

Code

An explicit slot for a coding-compatible endpoint

One request only

02 · FIT CHECK

Is Blossom Router useful for you?

Check what matches your workflow. The score is local to this page and is not stored.

0/5

Optional

A single provider may be simpler without routing.

I use both local and cloud modelsI want to reduce premium-model quota usageI repeat summaries, translations, logs, or small code tasksI prefer explicit rules over a hidden AI classifierMy providers expose OpenAI-compatible chat completions

03 · CONFIGURE

Build a practical first configuration.

Start with two useful tiers. Add strong and code only after real tasks justify them.

Local Ollama modelCloud base URLNormal cloud modelStrong cloud model (optional)

API keys stay in environment variables. This generator never asks for or stores a key.

router.yaml

default_route: normal

providers:
  local:
    base_url: http://localhost:11434/v1
    model: qwen3:8b
  normal:
    base_url: https://api.example.com/v1
    model: your-normal-model
    api_key_env: BLOSSOM_NORMAL_API_KEY
  strong:
    base_url: https://api.example.com/v1
    model: your-strong-model
    api_key_env: BLOSSOM_STRONG_API_KEY

LOCAL UI

Prefer buttons to terminal commands?

Run one command to open the local control panel. Choose a route, paste a task, preview the decision, and see the result without handling API keys in a webpage.

bloom serve

04 · ROADMAP

A deliberate path to 1.0.

Stabilize the small router first. Expand only when real usage proves the next requirement.

01Now

v0.1 · Stable preview

CLI, local UI, bilingual guides, safe configuration, packaging baseline

02Next

v0.2 · Daily usefulness

Config diagnostics, request timeouts, streaming, usage evaluation

03Then

v0.3 · Provider depth

Explicit adapters, cost guardrails, opt-in bounded fallback

04Goal

v1.0 · Trusted tool

Stable contracts, signed releases, clean-install acceptance, upgrade policy

05 · FAQ

Frequently asked questions

Does the public website call my models?

No. The public site only explains and generates configuration text. Run bloom serve for the local operating UI.

Does automatic routing consume tokens?

No. The MVP uses deterministic keyword and length rules. Dry-run never calls a provider.

Can it use Codex?

The code slot accepts a configurable OpenAI-compatible chat-completions endpoint. It does not assume a proprietary transport.

Where are API keys stored?

Use environment variables referenced by api_key_env. Keys are not compiled into bloom or exposed to the browser UI.

Which local model should I use?

Choose a Qwen or other Ollama model that fits your Mac memory. Use a model you already run reliably before chasing benchmarks.

What should stay out of this MVP?

RAG, MCP, databases, agent frameworks, hidden retries, and automatic fallback remain deferred until actual usage proves a need.