Jejoo beta Web UI

Optimize your AI agent

Jejoo analyzes execution traces to uncover unnecessary work, expensive execution paths, repeated context, retries, loops, and other sources to avoid cost and latency.

Analyze my agent

You can analyze traces on their own, but providing a complete workflow definition enables more precise analysis of the system’s structure and execution behavior. A workflow definition can be structured JSON describing nodes, edges, and routing logic, or source code that defines the workflow. Upload structured workflow JSON or the Python/JSON/YAML files that define your workflow.

Free during the beta. Run analyses at no cost for a limited time while Jejoo is in beta.

New analysis

Beta
Add files together or one at a time. Selecting the same filename replaces its previous selection. Upload one or more JSON/JSONL files containing native traces or operation records. JSON may be an array or an object with a records array. Include a run or trace ID on every record. Combined limit: 50 runs, 500 records, and 5 MiB.
Add files together or one at a time; selected files remain listed above. Selecting the same filename replaces its previous selection. Recommended for more precise structural analysis. Upload workflow JSON, or select up to 10 Python/JSON/YAML source files (200000 bytes of source combined). Include orchestration and called modules. Source files are read without execution and converted into a workflow definition.
A complete workflow description enables more precise analysis.
Edit this comma-separated list of field names to remove from the trace records. Removal is best effort; review your data for sensitive content before sending.

Your traces remain yours. We use them only to provide your Jejoo analysis and do not use them to train general-purpose AI models.

Do not upload passwords, API keys, payment card data, PHI, or data you are not authorized to process.

Before storage or analysis, the server screens trace records and workflow inputs for likely credentials, payment card numbers, direct personal identifiers, and health-data fields. Detected submissions are rejected. This screening is an additional safeguard, not a substitute for reviewing your data before upload.

By submitting, you agree to our Terms of Service and Privacy Policy and confirm that you are authorized to provide this data. Learn more about how we handle trace data.

Upload up to 50 complete runs and 500 records across all selected files. The browser merges the files but does not sample or drop records; it sends all supplied records after removing the selected fields, provided the request is at most 1 MiB. Every upload is processed server-side through the normalization model: a masked structural profile and examples are sent to the model, then its validated mapping is applied to the full upload. Workflow source files are also sent to the normalization model to extract nodes, edges, and source evidence. The investigation model can search and read bounded excerpts from those uploaded files to explain trace behavior; review them for sensitive content before upload. The extracted workflow, source hashes, and extraction warnings appear with the results. The uploaded records and canonical facts are stored server-side for analysis. Raw file contents are not saved in browser storage. An expiring private web session isolates your uploads, jobs, and results from other beta visitors.

Open an existing job

You can open jobs created by this browser’s current private session. Jobs from another browser, session, or API key are not accessible.