# Demo 01 — Zero to Movie Recommendation Agent in 5 Minutes **Platform aspects**: Agent creation, `ai init`, `ai serve`, `ai run`, `ai web` **Graph**: Neo4j Recommendations demo database (`recommendations` / `recommendations` / `recommendations`) **Audience**: New users; developer onboarding; conference opening slot --- ## The Scenario You've just heard about agent-intelligence.ai. You want to go from zero to a running AI agent that answers natural-language questions about movies — in under five minutes, on your laptop, with no cloud account required. The agent will use the Neo4j Movies demo database and a local Cypherlite graph for caching. --- ## Prerequisites ```bash # Install the ai CLI (one-line curl install) curl -fsSL https://agent-intelligence.ai/install.sh | sh # Verify ai --version # ai version 0.4.1 (commit a3f2c1d, built 2026-03-09) ``` You need an Anthropic API key: ```bash export ANTHROPIC_API_KEY=sk-ant-... ``` --- ## Step 1 — Create your agent with `ai init` ```bash ai init my-movie-agent cd my-movie-agent ``` The onboarding agent starts an interview: ``` ╔══════════════════════════════════════╗ ║ agent-intelligence.ai v0.4.1 ║ ║ Agent Onboarding ║ ╚══════════════════════════════════════╝ Hi! I'll help you configure your agent in a few questions. What should your agent be called? › movie-assistant What does it do? › Answer questions about movies, directors, and actors Do you have a Neo4j database to connect to? (y/n) › y Connection URI › neo4j+s://demo.neo4jlabs.com:7687 Username › recommendations Password › recommendations Database (leave blank for default) › recommendations ✓ Connected to Neo4j ✓ Introspecting schema... Node labels: Movie, User, Genre, Actor, Director Relationship types: RATED, IN_GENRE, ACTED_IN, DIRECTED Key properties: title, year, plot, imdbRating, tmdbId, name, userId Generating agent.toml and toolbox.yaml... ✓ Created agent.toml ✓ Created toolbox.yaml ✓ Created .env.example All done! Run `ai serve` to start your agent. ``` Look at what was generated: ```bash ai show ``` ```toml [agent] name = "movie-assistant" description = "Answer questions about movies, directors, and actors" system_prompt = """ You are a helpful movie expert with access to a comprehensive graph database of movies, actors, directors, and genres. You can answer questions about: - Movie details (release year, tagline, cast, director) - Actor filmographies and collaborations - Genre exploration and recommendations - "Six degrees of separation" style relationship paths Always cite specific movie titles, years, and actor names in your answers. """ [agent.model] provider = "anthropic" model = "claude-sonnet-4-6" api_key = "${ANTHROPIC_API_KEY}" [agent.budget] max_turns = 10 context_warn_ratio = 0.70 context_compact_ratio = 0.80 context_abort_ratio = 0.95 [agent.toolbox] endpoint = "http://localhost:15000/mcp/sse" transport = "sse" ``` The generated `toolbox.yaml` has Cypher tools pre-built from the schema: ```yaml sources: recs-db: kind: neo4j uri: neo4j+s://demo.neo4jlabs.com:7687 user: recommendations password: recommendations database: recommendations tools: find_movies_by_title: kind: neo4j-query source: recs-db description: Search for movies by title (partial match), returns rating and year parameters: - name: title type: string description: Movie title or partial title to search for query: | MATCH (m:Movie) WHERE toLower(m.title) CONTAINS toLower($title) RETURN m.title AS title, m.year AS year, m.plot AS plot, m.imdbRating AS imdb_rating ORDER BY m.imdbRating DESC LIMIT 10 find_movies_by_genre: kind: neo4j-query source: recs-db description: List top-rated movies in a specific genre parameters: - name: genre type: string description: Genre name (e.g. Action, Comedy, Drama, Thriller) query: | MATCH (m:Movie)-[:IN_GENRE]->(g:Genre {name: $genre}) RETURN m.title AS title, m.year AS year, m.imdbRating AS imdb_rating ORDER BY m.imdbRating DESC LIMIT 12 find_similar_movies: kind: neo4j-query source: recs-db description: Find movies similar to a given title based on shared actors and genres parameters: - name: title type: string query: | MATCH (m:Movie {title: $title})<-[:ACTED_IN]-(a:Actor)-[:ACTED_IN]->(rec:Movie) WHERE rec <> m WITH rec, count(a) AS overlap, collect(a.name) AS shared_actors MATCH (rec)-[:IN_GENRE]->(g:Genre)<-[:IN_GENRE]-(m) WITH rec, overlap, shared_actors, count(g) AS genre_overlap RETURN rec.title AS recommended, rec.year AS year, rec.imdbRating AS imdb_rating, overlap AS shared_actors_count, genre_overlap AS shared_genres ORDER BY (overlap * 2 + genre_overlap) DESC LIMIT 8 get_user_recommendations: kind: neo4j-query source: recs-db description: Get personalized recommendations for a user based on their ratings parameters: - name: user_id type: string description: User ID string (e.g. "564") query: | MATCH (u:User {userId: $user_id})-[r:RATED]->(m:Movie) WHERE r.rating >= 4.0 WITH u, collect(m) AS liked MATCH (m2:Movie)<-[:ACTED_IN]-(a:Actor)-[:ACTED_IN]->(liked_m:Movie) WHERE liked_m IN liked AND NOT (u)-[:RATED]->(m2) WITH m2, count(a) AS score RETURN m2.title AS recommended, m2.year AS year, m2.imdbRating AS imdb_rating, score AS relevance_score ORDER BY score DESC LIMIT 10 ``` --- ## Step 2 — Start the agent ```bash ai serve ``` ``` Starting agent-intelligence.ai... [mcp-toolbox] Starting subprocess on :15000... [mcp-toolbox] ✓ Health check passed (238ms) [mcp-toolbox] ✓ Loaded 4 tools from movies-db [a2a-server] Listening on :8080 [mcp-server] Listening on :8081 agent-card: http://localhost:8080/.well-known/agent.json a2a endpoint: http://localhost:8080/a2a mcp endpoint: http://localhost:8081/mcp (Streamable HTTP) mcp endpoint: http://localhost:8081/sse (legacy SSE) Ready. Press Ctrl+C to stop. ``` --- ## Step 3 — Run your first query In a second terminal: ```bash ai run "Who directed The Matrix, and what other movies did they direct?" ``` ``` Thinking... → find_directors_by_movie(title="The Matrix") ← director: "Lana Wachowski", movie: "The Matrix", released: 1999 → find_movies_by_actor(name="Lana Wachowski") ← [Cloud Atlas (2012), Jupiter Ascending (2015), The Matrix Revolutions (2003), ...] The Matrix was directed by **Lana Wachowski** (then credited as Larry Wachowski). After The Matrix, the Wachowski siblings went on to direct: - **The Matrix Reloaded** (2003) — the first sequel, expanding Neo's conflict with the machines - **The Matrix Revolutions** (2003) — the trilogy conclusion - **Speed Racer** (2008) — a visually bold adaptation of the anime series - **Cloud Atlas** (2012) — co-directed with Tom Tykwer; an intricate multi-timeline epic - **Jupiter Ascending** (2015) — original sci-fi space opera All Matrix films are in the graph. Cloud Atlas and Jupiter Ascending are also present. ``` Try a more complex query: ```bash ai run "Find me six degrees of separation between Tom Hanks and Keanu Reeves" ``` ``` Thinking... → find_movies_by_actor(name="Tom Hanks") ← [Cast Away, Forrest Gump, The Green Mile, Apollo 13, You've Got Mail, ...] → find_movies_by_actor(name="Keanu Reeves") ← [The Matrix, Speed, Johnny Mnemonic, Much Ado About Nothing, ...] → recommend_similar_movies(title="The Green Mile") ← shared actors/directors lead to other films... Here's a connection between **Tom Hanks** and **Keanu Reeves**: Tom Hanks → (appeared in) **The Da Vinci Code** → (also starred) **Audrey Tautou** Audrey Tautou → (appeared in) **Amélie** → (featured) ... Actually, a shorter path: both are in the Movies graph but have no direct co-starring credit in this dataset (the Neo4j demo graph has ~3,600 nodes). Within the available data, the closest connection is through directors who worked with both actors' peers. Want me to look for a specific pathway, or explore what films they share genre-overlap with? ``` --- ## Step 4 — Open the web console ```bash ai web ``` ``` Starting web console... Connecting to agent at http://localhost:8080... ✓ Agent online (movie-assistant v0.4.1) Web console: http://localhost:8888 Opening browser... ``` The browser opens to a chat interface where you can: - Type natural language queries - See tool calls and results in the debug panel (`--debug` flag) - Save interesting conversations as eval cases ```bash # With debug panel enabled: ai web --debug ``` In the debug panel you'll see each tool call expand inline: ``` ▶ find_movies_by_actor {"name": "Tom Hanks"} 32ms ▶ find_movies_by_actor {"name": "Keanu Reeves"} 28ms tokens: 847 in / 312 out ($0.0009) ``` --- ## What you just built In five minutes and one command, you created an agent that: - **Speaks natural language** — no Cypher knowledge required from the user - **Uses a real graph database** — 3,682 nodes, 9,629 relationships - **Explains its reasoning** — tool calls are visible via `ai web --debug` or `ai run --verbose` - **Serves over two protocols** — A2A on `:8080`, MCP on `:8081` - **Has a web UI** — shareable chat interface at `:8888` Next: [Demo 02 — Company Intelligence with GraphRAG](demo-02-company-intelligence.md) --- ## Appendix — Demo fixture files **`toolbox.yaml`** (full generated config): [`fixtures/demo-01/toolbox.yaml`](fixtures/demo-01/toolbox.yaml) **`agent.toml`** (full config): [`fixtures/demo-01/agent.toml`](fixtures/demo-01/agent.toml) ```bash # Quick-start with fixture files (skip ai init): curl -fsSL https://agent-intelligence.ai/downloads/demo-01-fixtures.tar.gz | tar xz ai serve ```