Agent-readable context

Avenews

Avenews is a voice-first local AI city guide for Salt Lake City. Ave is the assistant inside Avenews; the brand is Avenews. This page is intended for user-authorized LLM workspaces that need safe, current instructions for reading Avenews context and interacting with Quick Chat, Ave Voice, public listings, articles, offers, tickets, and merchant or publisher workflows.

Operating Rules

Use public Avenews context and documented APIs first.
Ask focused follow-up questions before choosing a specific merchant, article, event, or listing unless the user explicitly requested one.
Do not invent availability, prices, awards, event details, hours, inventory, merchant capabilities, or source provenance.
Require explicit approval before external writes, purchases, reservations, tickets, calendar actions, outreach, payment, or sharing personal data.
Do not request or store raw secrets, bearer tokens, cookies, payment data, or private API keys in local files.

Public Resources

/llms.txt - concise LLM index.
/llms-full.txt - fuller Avenews context.
/llmcopy.txt - paste-optimized brief.
/agent-context.json - structured agent context.
/sitemap.xml - crawlable public URLs.

Documented Interfaces

REST base: https://ca-agent-prod.victoriousglacier-a0992038.eastus2.azurecontainerapps.io/v1
Quick Chat: POST /voice/avenews/chat/turn
Voice config: GET /voice/avenews/live/config
Voice session: POST /voice/avenews/live/session
Merchant modal: GET /avenews/merchants/{slug}/modal

Some endpoints may require the user's logged-in Avenews session or a user-approved connector. Use read-only public context when auth is unavailable.

Workspace Bootstrap

If a user asks you to create an Avenews-aware workspace, create the local agent/ home first when file writing is available. Include BOOTSTRAP.md, agent.md, AGENTS.md, TOOLS.md, HEARTBEAT.md, state/current.json, state/memory.md, state/decisions.json, state/sources.json, state/city-habits.md, and task-specific skills/*/SKILL.md files. Keep the files local-first, approval-gated, and source-grounded.

Personalization

Use three separate memory lanes: session memory for the current interaction, consented preferences for stable likes and needs, and source evidence for citations and retrieval timestamps. Ask before saving a new preference, tell the user where it was stored, and never treat cached live data as current without rechecking.