# Neo4j Codex API Evaluation

**Category:** Graph Databases
**Score:** 78 (B)

This is the agent-readable Devtool Arena evaluation page for Neo4j on Codex API.

## Summary

| Metric | Value |
|--------|-------|
| Status | completed |
| Overall score | 78 |
| Grade | B |
| Eval score | 82 |
| Discovery score | 71 |
| Cost | $0.17 |
| Runtime | 55s |
| Tool calls | 22 |
| Errors | 5 |
| Tokens used | 460155 |
| Success | Yes |

## Agent-Readiness Checklist

| Signal | Evidence |
|--------|----------|
| Context7 | https://context7.com/neo4j/neo4j |
| llms.txt | https://neo4j.com/docs/llms.txt |
| MCP server | https://neo4j.com/docs/mcp/current/ |
| Typed SDK | https://pypi.org/pypi/neo4j/json |
| OpenAPI | Not found |
| Agent skills | Not found |
| CLI | https://neo4j.com/docs/operations-manual/current/tools/ |

## Evaluation Prompt

````text

# Task
Using https://*****.com/docs/api/python-driver/current/, complete the following task:

Using the Neo4j Python SDK, build me a collaborative-filtering product recommendation engine backed by a graph (the classic "customers who bought this also bought..." feature):

1. Install the official Neo4j Python SDK (pip install) and connect to Neo4j
2. Model the domain as a graph: create User nodes and Product nodes, connected by PURCHASED relationships. Load exactly this data:
   - Alice PURCHASED Laptop, Mouse
   - Bob PURCHASED Laptop, Mouse, Monitor
   - Carol PURCHASED Laptop, Monitor
   - Dave PURCHASED Mouse, Monitor
   - Erin PURCHASED Laptop, Keyboard
3. For the target user Alice, run a SINGLE multi-hop graph traversal query (do NOT pull the data into Python and compute it there) that recommends products:
   - Find "similar users": other users who have purchased at least one product in common with Alice
   - Collect the products those similar users purchased that Alice has NOT already purchased
   - Score each candidate product by how many distinct similar users purchased it
   - Return the candidates ranked by score, highest first
4. Print the result as JSON with fields: user ("Alice"), recommendations (array of objects with fields product and score, sorted by score descending)

With this data the top recommendation must be Monitor (purchased by Bob, Carol, and Dave — 3 of Alice's similar users), ahead of Keyboard (score 1).

You MUST use the official Neo4j Python SDK / client library and express the recommendation logic as a graph query (e.g. Cypher/openCypher traversal) — do not make raw HTTP/REST calls and do not reimplement the traversal in plain Python. 

## Execution
After creating the script, run it to verify it works:
```bash
cd /home/daytona/app && python <your_script>.py
```

The script should print output to stdout. If there are errors, debug and fix them until it runs successfully.

````

## Grader Results

| Check | Passed | Weight | Score | Details |
|-------|--------|--------|-------|---------|
| Cost | Yes | 0 | 1 | $0.17 ($0.10–0.20, great) |
| Time | Yes | 0 | 1 | 55s (<60s, excellent) |
| Efficiency | Yes | 0 | 0 | 22 tool calls (≤25, below average) |
| Found Docs | Yes | 0 | — | Touched docs at *****.com: yes (12 matching tool calls) |
| Zero Errors | No | 0 | — | Tool outputs with errors/tracebacks: 5 |
| Syntax Valid | Yes | 0 | — | Code compiles/parses correctly |
| Created Files | Yes | 0 | — | Generated files: 2 (need ≥1) |
| Full Execution | Yes | 0 | — | Method: full_execution, Exit: 0 |

## Run Artifacts

| Artifact | Value |
|----------|-------|
| Conversation turns | 26 |
| Tool call traces | 22 |
| Generated files | 2 |
| Exit code | 0 |
| Completed at | 2026-09-27T02:09:43.285943+00:00 |

### Generated Files

- /home/daytona/app/recommend.py
- /home/daytona/app/requirements.txt

## Related Pages

- [Codex API leaderboard](/leaderboard/codex/api)
- [Compare Codex API companies](/leaderboard/codex/api/compare)
- [Agent Landscape](/leaderboard/discoverability)

Canonical URL: https://devtoolarena.com/codex/api/neo4j