Recipes

Opinionated end‑to‑end examples that compose Providers, Models, Agents, Tools (MCP), Prompt Studio, Embeddings, and Schedules.

A) “Web News → Daily Email” agent

Goal

Aggregate fresh finance headlines, summarize concisely with links, and send a daily email.

Steps

1. Providers: add OpenAI (or your preferred fast/cheap model provider).
2. Models: save a low‑temp chat model as gpt-4.1-mini-factual.
3. MCP: ensure webtools (or rss) is connected and exposes a function like get_feed_headlines.
4. Agents: create web-news-finance with a system prompt that:
   • instructs use of the get_feed_headlines tool,
   • requests accurate links and short summaries,
   • defines selection criteria (accuracy, novelty).
5. Prompt Studio: pick the agent, turn Streaming ON, verify tool calls/outputs; refine the prompt.
6. Schedules: run daily at 07:45 (your TZ); add recipients for delivery.

API parity

• Run via chat: POST /v1/chat/completions with model: "web-news-finance" — see Chat Completions
• Tools inventory: GET /v1/core/tools — see Core Utilities

---

B) “GitHub Triage” agent

Goal

Search repos for new issues, post comments, and optionally create tracking issues.

Steps

1. MCP: connect the github server with a least‑privilege PAT.
2. Agents: create code-reviewer bound to a deterministic model; enable Tools; grant search_issues, create_issue, add_issue_comment.
3. Prompt Studio: test with a repo/org filter; confirm tool traces before enabling write actions.
4. (Optional) Schedules: run hourly and email a digest of new issues found by search.

API parity

• Tools inventory: GET /v1/core/tools — see Core Utilities
• Run via chat: POST /v1/chat/completions with model: "code-reviewer" — see Chat Completions