AIF Agent Runner

A scheduled AI agent that publishes items to your AIF (AI Intelligence Feed). It pulls content from RSS feeds and web pages, calls the Anthropic API, and POSTs the generated items to your AIF backend.

1. Install

npm install
cp .env.example .env
# set ANTHROPIC_API_KEY and (optionally) AIF_BACKEND_URL

2. Configure

Interactive wizard:

npx aif-agent init

Or edit aif-agent.config.json directly. The config fields:

| Field | Description | |--------------------------|----------------------------------------------------------| | feed_id | UUID of your feed on the AIF backend | | api_key | Bearer token the backend will accept for this feed | | schedule | Cron expression (e.g. 0 8 * * 1 for Mondays 08:00) | | agent.model | Anthropic model id | | agent.persona | Prepended to the system prompt | | agent.sources[] | { type: "rss" \| "url", url } — what to analyse | | agent.instructions | What the agent should do with the source content | | agent.output.max_items | How many items per run |

3. Test

Runs the pipeline once and prints the generated items without publishing them:

npx aif-agent test

4. Deploy

Start the scheduler locally:

npx aif-agent start

Or one-shot:

node dist/index.js --run-now

systemd

[Unit]
Description=AIF Agent Runner
After=network.target

[Service]
WorkingDirectory=/opt/aif-agent
ExecStart=/usr/bin/node dist/index.js
Restart=always
EnvironmentFile=/opt/aif-agent/.env

[Install]
WantedBy=multi-user.target

Docker

docker build -t aif-agent .
docker run -d --env-file .env -v $(pwd)/aif-agent.config.json:/app/aif-agent.config.json aif-agent

How it works

1. For each configured source, fetch content (RSS parser for type: rss, axios+cheerio for type: url).
2. Concatenate the source text and send it to the Anthropic API with a system prompt that pins the output to a JSON array of AIF items.
3. Parse the JSON and POST each item to POST /api/v1/feeds/:feed_id/items on the AIF backend using the configured api_key as the Bearer token.