Multi-stage problem solving via DeepTutor (plan → reason → write).
Use the exec tool to invoke DeepTutor's multi-stage solver for complex math, science, or engineering problems.
deeptutor run deep_solve "<problem description>" --format json -l <lang>
| Flag | Description |
|---|---|
-l <lang> | Response language: en or zh |
-t rag | Enable RAG to ground in knowledge base |
-t web_search | Enable web search |
-t code_execution | Enable code verification |
-t reason | Enable dedicated reasoning |
--kb <name> | Knowledge base to use with RAG |
--config detailed_answer=false | Return a concise answer |
Prove a theorem:
deeptutor run deep_solve "Prove that for any positive integer n, n³ - n is divisible by 6" --format json -l en
Solve with knowledge base:
deeptutor run deep_solve "Derive the Euler-Lagrange equation" --format json -l en -t rag --kb physics-textbook
Solve with code verification:
deeptutor run deep_solve "Find all roots of x⁴ - 5x² + 4 = 0" --format json -l en -t code_execution
timeout=300 with the exec tool.--format json flag outputs NDJSON stream events. Parse the content field from lines with "type": "content" to get the final answer. Concatenate all content events for the full response.