aohp-os/aohp/aohp-uda

AOHP User Defined App (UDAGen)

Use this skill when the user wants a personalized mini-app (UDA / PUI) generated on the AOHP device.

Pair with `uda-app` after generation completes (hasApp: true) to install, pin, and launch.

Prerequisites

1. LLM credentials on device (not baked into the image):

``bash aohp uda config-set \ --api-key "$UDA_API_KEY" \ --model "doubao-seed-2-0-pro-260215" \ --base-url "https://ark.cn-beijing.volces.com/api/v3" \ --provider openai ``

1. The uda Alpine container is created automatically on first run.

Choose a workflow

A. Idea-only (fastest)

No custom input files. UDAGen auto-drafts mock API docs from --idea.

aohp uda generate \
  --app-name "Google Search" \
  --idea "Google Search mini-app: search arbitrary keywords on a single mobile page; show Google results after submit"

Note the returned jobId, then poll aohp uda status -j <jobId> until status: completed and hasApp: true.

B. Structured input (richer apps)

Stage per-job input before uda.generate. When /opt/udagen/workspace/<jobId>/input/ contains supported files, generation uses that tree instead of the empty template.

1. Allocate input scaffold

aohp uda input init
# → { "jobId": "job-1710000000000", "containerInputDir": "...", "hostInputDir": "..." }

2. Write input files (repeat as needed)

JOB=job-1710000000000
 
aohp uda input write -j "$JOB" --path requirements/idea.md --content "$(cat <<'EOF'
# Google Search App
Single search field; submit runs a Google search for arbitrary keywords.
Mobile-first; pin-friendly launcher name: Google Search.
EOF
)"
 
aohp uda input write -j "$JOB" --path app/manifest.json --content '{
  "name": "Google Search",
  "short_name": "Search",
  "theme_color": "#4285F4",
  "background_color": "#FFFFFF",
  "description": "Search Google with any keyword"
}'
 
aohp uda input write -j "$JOB" --path mock_docs/openapi.yaml --content "$(cat <<'EOF'
openapi: 3.0.3
info:
  title: Google Search Proxy
  version: 1.0.0
paths:
  /search:
    get:
      summary: Search by keyword
      parameters:
        - name: q
          in: query
          required: true
          schema: { type: string }
      responses:
        "200":
          description: Redirect or embed URL for Google results
EOF
)"

3. Generate with the same job id

aohp uda generate -j "$JOB" \
  --app-name "Google Search" \
  --idea "Google Search: arbitrary keywords"

Alternative: edit files via sandbox:

aohp sandbox exec uda -- ls /opt/udagen/workspace/$JOB/input/

Host bind path: /data/aohp/shared/uda/<jobId>/input/

After generation

aohp uda status -j <jobId>    # poll until hasApp
aohp uda install -j <jobId> --pin
aohp uda launch -j <jobId>

See skill `uda-app` for install/launch details.

Input bundle reference

Supported extensions: .md, .txt, .json, .yaml, .yml, .csv, .tsv

Baked template (device): /opt/udagen/template-input/ Per-job input: /opt/udagen/workspace/<jobId>/input/

RPC mapping

uda.generate optional params: jobId, appName, idea, inputDir (container path override).

Output layout

/data/aohp/shared/uda/<jobId>/ (container: /opt/udagen/workspace/<jobId>/):

• app/index.html — WebView app
• mock/server.py — local API on 127.0.0.1:8787
• prd.json, design_spec.json, udagen.log — pipeline artifacts

Tips

• Generation runs in the background inside uda; poll uda.status (minutes).
• Do not commit API keys into the image; use uda config-set.
• For idea-only tasks, workflow A is enough; use B when the user supplies API shapes, data samples, or detailed requirements files.