Arabian Sand Boa: Python interpreter with frontier intelligence conditional eval

Arabian Sand Boa is a Python script that rewrites if/elif conditions at the AST level, sending string conditions to an LLM for evaluation. The tool allows natural-language clauses to control program flow, but sends local variables to the LLM endpoint and may be slow for loops.

Run a Python file where string if conditions are decided by an LLM . arabian sand boa is a single, dependency-free script standard library only . It reads your target file, rewrites every if / elif at the AST level, and runs it. When a condition resolves to a non-empty string , the string is treated as a natural-language clause and sent to an LLM along with the variables currently in scope; the model's True / False answer decides which branch runs. Every other condition resolves the normal way via bool . ./arabian sand boa <file.py args... it's executable --debug or -d prints, to stderr, the exact prompt sent to the LLM and the raw reply for each clause it evaluates. The target file runs with name == " main " , so its main block executes. The LLM endpoint is configured entirely through environment variables, read lazily — a target that only uses ordinary non-string conditions never needs them. | Variable | Meaning | |---|---| BOA LLM URL | Full chat-completions endpoint URL | BOA LLM API KEY | Bearer token for that endpoint | BOA LLM MODEL | Model name to request | The endpoint is expected to speak the OpenAI-style chat-completions protocol POST a JSON body with model and messages , get back choices 0 .message.content . Copy .example.env to .env , fill in your values, and load it: cp .example.env .env edit .env set -a; source .env; set +a python3 arabian sand boa example.py .env is gitignored; .example.env is the tracked template. If any required variable is missing when an LLM call is needed, the run fails with a clear error naming the missing variable s . Rewrite. An ast.NodeTransformer replaces each if <test : with if if hook <test , "<source text of test " : . Decide. if hook evaluates <test . If it's a non-empty string, the hook gathers the caller's non-dunder local variables and asks the LLM whether the clause holds. Otherwise it returns bool <test and no API call is made. Parse. The reply is lower-cased and stripped: true → take the branch, false → skip it, anything else → take the branch cautious default . Your local variables leave the machine. For every string condition, the in-scope non-dunder local variables are serialized and sent to the configured LLM endpoint. Do not run this over sensitive data.- An LLM call fires for every executed string condition , so loops with string clauses are slow and chatty against the endpoint. - Decisions are only as reliable as the model and the wording of your clause, and may vary between runs. See example.py for a runnable showcase mixing natural-language clauses drinking age, admin privileges, account health with an ordinary boolean condition.