Code Mode for AI agents

One program instead of many calls

1. 1
   The agent reads the library

   code_mode_reference gives the waiso library’s calls for the language it picks — every language Waiso runs.

2. 2
   It writes one program

   Reads and writes alike: list, filter, change, return a summary.

3. 3
   Waiso runs it

   In the project’s sandbox (no network), as the token: every call is checked against the token’s scope and its owner’s access.

4. 4
   Only the outcome comes back

   The program’s result, its logs and writes — every change it made, in order, even when it stops halfway.

There is nothing to turn on: Code Mode is two tools of the same MCP server, and the server’s own guide tells agents to prefer them for any job of several steps. You can still ask for it in words — “use Code Mode” — and the agent writes a program like this one.

import waiso

def main():
    paused = []
    for flow in waiso.api.flows.list()["flows"]:
        runs = waiso.api.runs.list(target=f"flow:{flow['id']}", status="failed", limit=50)
        if runs["total"] > 3:
            for s in waiso.api.schedules.list()["schedules"]:
                if s["target"].get("flowId") == flow["id"] and s["enabled"]:
                    waiso.api.schedules.update(s["id"], enabled=False)
                    paused.append(s["name"])
    return {"paused": paused}

When to use it, and when not

Code Mode pays off when a request means looking at many things and deciding: clean up runs, pause what keeps failing, rename a batch of scripts, collect a report across flows. The agent spends its tokens on the program and on the short answer, not on every list it would otherwise read. For a single action — run this script, show this flow — the MCP tool for it is just as good and easier to follow.

The program can only reach the platform, and only as the token: it has no network, no files outside its run, and a call outside the token’s scope raises the library’s error. A program that loops forever is stopped at its time limit; what it changed until then is still listed.

Limits and what you see

Every Code Mode run is in Runs as Code Mode with the token’s name; open it to see the program, the result and a Changes made table. Each change is also in the audit log, with the token and the run.