# Daily Brief — Implementation Notes

## Architecture: context-from Chaining

```
┌──────────────────────────────┐
│  cron: 0 15 * * *            │
│  script: collector.sh        │  ← no-agent, free public APIs
├──────────────────────────────┤
│  Script stdout → agent       │  ← injected as prompt context
│  context                     │
├──────────────────────────────┤
│  Agent: curates → formats    │  ← lightweight LLM pass only
│  → delivers to Telegram      │
└──────────────────────────────┘
```

## Cost Breakdown

| Stage | Token Cost | Frequency |
|-------|-----------|-----------|
| Script fetch (shell) | $0 | 1x per tick |
| Agent curation | Full model | 1x per tick |

vs. old approach that relied on web_search + x_search (both failing silently).

## Sources (Free, No API Key)

| Source | Endpoint | Reliability |
|--------|----------|-------------|
| Weather | `wttr.in/Placerville,CA` | 99.9% |
| Hacker News | Firebase API (v0) | 99.9% |
| GitHub Trending | `api.github.com/search/repositories` | 99% (unauthed rate limit: 60/hr) |

## Previous State (Before June 20)

Old prompt instructed the agent to use web_search and x_search tools. Both were failing — Firecrawl unconfigured, xAI credits exhausted. Brief returned `[SILENT]` for weeks.

## Key Insight

The script + context_from pattern turned a dead cron job into a working one in <5 minutes with zero ongoing API costs.

## Cron Job (Production)

```bash
hermes cron update e94447937052 \
  --script daily-brief-collector.sh \
  --prompt "Curate from the injected context. Output max 8 bullets. Silent if no data."
```

- **Job ID:** e94447937052
- **Schedule:** `0 15 * * *` (3pm UTC = 8am PDT)
- **Delivery:** Telegram (here)
- **Model:** grok-4.3 via xai-oauth