Fix base_url handling for openhands/ provider with explicit None value

[xingyaoww]

Problem

The openhands/ provider was failing when base_url was explicitly set to None, which broke the new user setup flow in the CLI. This occurs when CLI saves settings to JSON with base_url=null and then reloads them.

Related issues:

• CLI PR chore: Merge test_editor_comprehensive with test_basic_operations #168 - Reported the breaking change
• CLI PR Refactor visualizer to use modular approach with individual event classes & consistent color #171 - Release PR held pending this fix

Root Cause

PR #1273 changed the openhands/ provider conversion logic in llm.py:

# Before (always set):
d["base_url"] = "https://llm-proxy.app.all-hands.dev/"

# After (broke with explicit None):
d["base_url"] = d.get("base_url", "https://llm-proxy.app.all-hands.dev/")

The Issue: Python's dict.get(key, default) only returns the default when the key is missing, not when it exists with value None. When CLI serializes config to JSON with base_url: null and deserializes it, the dictionary contains {"base_url": None}, causing dict.get() to return None instead of the default.

The Workaround in 1.4.0: Commit 54d93bc changed the test workflow to bypass the openhands/ provider entirely by using litellm_proxy/ directly with a hardcoded URL.

Solution

1. Fixed llm.py: Changed to use d.get("base_url") or default pattern to handle explicit None values correctly
2. Reverted workflow: Changed .github/workflows/run-examples.yml back to using openhands/ provider as intended
3. Added tests: Comprehensive test coverage for all scenarios

Code Changes

# Use `or` operator to handle both missing keys and None values
d["base_url"] = d.get("base_url") or "https://llm-proxy.app.all-hands.dev/"

This correctly handles:

• ✅ Key missing: d.get("base_url") → None → or returns default
• ✅ Key exists with None: d.get("base_url") → None → or returns default
• ✅ Key exists with value: d.get("base_url") → value → or short-circuits

Testing

Added comprehensive test cases in test_llm.py:

• test_base_url_for_openhands_provider - Existing test for missing base_url
• test_base_url_for_openhands_provider_with_explicit_none - New: Reproduces CLI bug with explicit None
• test_base_url_for_openhands_provider_with_custom_url - New: Ensures custom URLs still work

All tests pass:

• ✅ 467 LLM tests passed
• ✅ Pre-commit hooks passed
• ✅ Manual testing confirms fix works for all scenarios

Impact

Before Fix

• ❌ CLI users couldn't use openhands/ provider (settings saved with base_url=null)
• ❌ New user setup flow was broken
• ⚠️ Workaround required using litellm_proxy/ directly

After Fix

• ✅ CLI users can use openhands/ provider seamlessly
• ✅ New user setup flow works correctly
• ✅ Custom base_url override still works
• ✅ Test suite covers all scenarios

Checklist

• Tests added/updated
• Pre-commit hooks pass
• All existing tests pass
• Changes are backward compatible

@xingyaoww can click here to continue refining the PR

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	eclipse-temurin:17-jdk	Link
python	amd64, arm64	nikolaik/python-nodejs:python3.12-nodejs22	Link
golang	amd64, arm64	golang:1.21-bookworm	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:7d075e5-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-7d075e5-python \
  ghcr.io/openhands/agent-server:7d075e5-python

All tags pushed for this build

ghcr.io/openhands/agent-server:7d075e5-golang-amd64
ghcr.io/openhands/agent-server:7d075e5-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:7d075e5-golang-arm64
ghcr.io/openhands/agent-server:7d075e5-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:7d075e5-java-amd64
ghcr.io/openhands/agent-server:7d075e5-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:7d075e5-java-arm64
ghcr.io/openhands/agent-server:7d075e5-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:7d075e5-python-amd64
ghcr.io/openhands/agent-server:7d075e5-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-amd64
ghcr.io/openhands/agent-server:7d075e5-python-arm64
ghcr.io/openhands/agent-server:7d075e5-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-arm64
ghcr.io/openhands/agent-server:7d075e5-golang
ghcr.io/openhands/agent-server:7d075e5-java
ghcr.io/openhands/agent-server:7d075e5-python

About Multi-Architecture Support

• Each variant tag (e.g., 7d075e5-python) is a multi-arch manifest supporting both amd64 and arm64
• Docker automatically pulls the correct architecture for your platform
• Individual architecture tags (e.g., 7d075e5-python-amd64) are also available if needed

[enyst]

Suggested change

LLM_BASE_URL: ${{ secrets.LLM_BASE_URL }}

TBH this line is strange, I think maybe we don't actually need it?

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
openhands-sdk/openhands/sdk/llm
llm.py	416	153	63%	357, 362, 366, 370–371, 374, 378–379, 390–391, 393–394, 398, 415, 475, 505–507, 528, 532, 547, 553–554, 578–579, 589, 614–619, 640–641, 644, 648, 660, 665–668, 675, 678, 686–691, 695–698, 700, 713, 717–719, 721–722, 727–728, 730, 737, 740–745, 802–807, 864–865, 868–871, 913, 930, 984, 987, 990–998, 1002–1004, 1007, 1010–1012, 1019–1020, 1029, 1036–1038, 1042, 1044–1049, 1051–1068, 1071–1075, 1077–1078, 1084–1093, 1106, 1120, 1125
TOTAL	12425	5644	54%

[enyst]

Thank you!