fix(helpers): give the Atoms REST helpers a request timeout

[CaptainAni187]

requests applies no timeout unless you pass one, so a call without timeout= blocks until the OS gives up on the socket, which on a half-open connection can be many minutes. The Atoms REST helpers make 31 calls and 28 of them had none:

# call.py
response = requests.get(url, headers=self._get_headers())

# kb.py
response = requests.post(url, headers=self._get_headers(), json=payload)

AgentTools, the fifth helper in the same package, already gets this right. It takes a keyword-only request_timeout: float = 30.0, stores it, and passes it to every call:

def _get(self, path: str) -> Any:
    resp = requests.get(f"{self.base_url}/{path}", headers=self._headers(), timeout=self.request_timeout)

So this is Audience, CallAnalytics, Campaign and KB catching up with the one helper beside them that already does it.

It matters most where these helpers are actually used. A voice agent reading call analytics or a knowledge base inside a request path has no way to bound the wait today, and the generated client next to it enforces 60s by default, so the hand-written helpers are the one place where a hang can happen.

The change

The same keyword-only request_timeout param, the same 30.0 default, stored on the instance and passed to all 28 call sites. Nothing else moves: no new module, no shared constant to import, no change to any existing argument, and the param is keyword-only so no positional call breaks.

versioning.py and _envelope.py make no HTTP calls, so they are untouched.

Tests

tests/custom/test_helpers_request_timeout.py walks every public method on each of the four managers with requests replaced by a recorder, then asserts that no recorded call went out without a timeout. Driving it by reflection rather than naming 28 methods means a helper method added later is covered the day it lands.

Twelve cases: every call carries a timeout, a caller-supplied value reaches all of them, and the default equals the one AgentTools already uses so the two cannot drift. All twelve fail on main.

src/smallestai/atoms/helpers/** is .fernignored, so a regen keeps this.

Worth saying: 30.0 matches AgentTools, not the generated client's 60. I took the sibling helper as the closer precedent, but happy to move all five to 60 if you would rather they match the client.

[crim-app]

crim doesn't review pull requests automatically here.

Comment crim review on this pull request whenever you want a review.

[CaptainAni187]

crim review

[crim-app]

crim is reviewing this pull request. Findings will be posted shortly.

[crim-app]

LGTM

One-sentence assessment: A correct, backward-compatible fix that closes real hang-forever risk by giving every Atoms REST helper call a request timeout.

What this PR does

Adds a keyword-only request_timeout parameter (default DEFAULT_REQUEST_TIMEOUT = 30.0) to the Audience, CallAnalytics, Campaign, and KB constructors and passes timeout=self.request_timeout to every requests call in those helpers. This matches the pattern already used by AgentTools, whose 30.0 default the new tests assert against. A parametrized test suite verifies every helper request carries a timeout, honors a caller-supplied value, and that the default matches AgentTools.

Findings

No issues found.

• The new parameter is keyword-only with a default, so existing positional callers (base_url, api_key) are unaffected.
• requests timeout is per-connect/per-read, not a total deadline, so the 30s default does not prematurely abort large multipart uploads in kb.upload_document / audience.create that keep making progress.
• Default value is kept in sync with AgentTools via an explicit test.

[crim-app]

Good to merge

Every request finally wears a watch instead of waiting forever on a half-open socket, and there's a test that fails if anyone forgets to strap one on. Genuinely tidy work.

[abhishekmishragithub]

Superseded by #120 (re-homed onto an upstream branch so CI could run; your commit is included with authorship preserved, and the test type-check gaps for #112/#118 were fixed there). Shipping in 5.12.1.