Commit Graph
3 Commits
Author SHA1 Message Date
RosenTomovandAlexandre Teixeira c46d37d876 test(tool_execution): stop two tests leaking src.tool_execution into the suite (#2686)
* Make in-venv pip-fallback test independent of the runner's environment

test_pip_install_fallback_chain_propagates_failure_in_venv simulated the in-venv case by probing the real interpreter (sys.prefix != sys.base_prefix). That assumes the test runner is itself inside a venv. CI runs pytest with no venv, so venv_check reported not-in-venv, the negated guard flipped, the --user branch fired, and the assertion failed. Make venv_check exit 0 directly to simulate the in-venv condition deterministically, mirroring the outside-venv companion test.

* Stop agent-tool import shims from leaking into the admin-gate test

test_function_call_non_object_args and test_unknown_tool_calls stub heavy DB/auth deps at import time to load the real agent-tool stack, but they popped src.tool_execution and left core.auth stubbed without restoring. Popping and re-importing src.tool_execution rebinds the src package's tool_execution attribute, so test_edit_file's later 'import src.tool_execution as te' resolved to a different module object than the one execute_tool_block lives in. The monkeypatch on _owner_is_admin then missed, the non-admin edit_file gate never fired, and the edit went through (exit_code 0). Stop touching src.tool_execution and restore the heavy stubs after import. Verified the full suite is green on Linux (Python 3.11, matching CI).

---------

Co-authored-by: Alexandre Teixeira <111787685+alteixeira20@users.noreply.github.com>
2026-06-09 16:35:10 +01:00
RosenTomov 37356d8e3e Discover LM Studio via host/port scanning and native-API fingerprint (#1126)
Scan port 1234 and any custom port from LM_STUDIO_URL, add the LM_STUDIO_URL host to the discovery sweep alongside the Ollama env vars, and tag each discovered endpoint with its provider by fingerprinting the native /api/v1/models response (entries carrying key + architecture). Documents LM_STUDIO_URL in .env.example.
2026-06-02 23:04:58 +09:00
RosenTomov a493fb49b0 Use LM Studio-reported vision capability for image passthrough (#1130)
Read a model's capabilities.vision flag from LM Studio's native /api/v1/models so vision finetunes whose names lack a vision keyword still receive images, falling back to the name heuristic when the endpoint doesn't report it. The probe is short-TTL cached and restricted to local/LAN hosts, so remote/cloud endpoints are never contacted.
2026-06-02 23:01:04 +09:00