fix(agent-server): propagate load_memory preference to all launch paths

[vnktadithya]

HUMAN:

Resolves #4542 - The memory preferences weren't propagating through two of the three existing paths (agent and agent_settings).

---

AGENT:

Why

agent_settings.agent_context.load_memory is a global user preference, but it was only ever applied when a conversation launched via agent_profile_id. Conversations started with a concrete agent or an agent_settings payload silently lost the setting — no error, no warning, just a missing <MEMORY_CONTEXT> block, while MemorySection rendered its other variant so the transcript still looked healthy.

Root cause: _resolve_agent_from_profile was the only place the stored preference got stamped onto the agent, and that function is only reachable via the agent_profile_id branch. The other two request shapes (agent, and agent_settings — converted to agent by a mode="before" validator before _start_conversation even runs) never touched it.

Summary

• Hoisted the settings-store read above the agent_profile_id branch in _start_conversation, so the load_memory stamp applies once, after all three request shapes converge on request.agent, instead of only on the profile path.
• Added _with_load_memory() next to _append_system_message_suffix, using the same two-level model_copy pattern (required because AgentBase is frozen).
• Removed the now-dead load_memory parameter from _resolve_agent_from_profile (signature, docstring, and inline body).

Issue Number

Fixes #4542

How to Test

Unit tests:

uv run pytest tests/agent_server/test_agent_profile_conv_start.py -q

Result: All 39 passed.

New coverage added:

• test_direct_agent_launch_inherits_the_stored_memory_preference
• test_direct_agent_launch_leaves_memory_off_without_the_preference (parametrized)
• test_agent_settings_launch_inherits_the_stored_memory_preference
• test_agent_settings_launch_leaves_memory_off_without_the_preference (parametrized)
• test_agent_launch_preserves_context_when_load_memory_already_true

Existing test_profile_launch_inherits_the_stored_memory_preference and test_profile_launch_leaves_memory_off_without_the_preference pass unchanged — they now exercise the post-branch stamp instead of the in-helper one.

Unit tests alone rely on mocked stores, so I also ran an end-to-end pass against a real (non-mocked) settings store and a real MEMORY.md:

uv run python .pr/verify_load_memory_all_paths.py

Output:

Persisted load_memory=True for real at C:\temp\memory-check-settings\settings.json

=== Part A: does load_memory() read the file back? ===
# Project memory (.openhands/memory/MEMORY.md)
# Verification note
load_memory propagated correctly if you can read this.
PASS

=== Part B: does the real settings store reach the launched agent? ===
agent  -> agent_context.load_memory = True
agent_settings  -> agent_context.load_memory = True
All checks passed.

Acceptance criteria (from #4542)

• With load_memory: true stored, a conversation started with a concrete agent has agent_context.load_memory is True — test_direct_agent_launch_inherits_the_stored_memory_preference
• Same for an agent_settings payload — test_agent_settings_launch_inherits_the_stored_memory_preference
• Same for agent_profile_id (existing behaviour preserved) — test_profile_launch_inherits_the_stored_memory_preference (unchanged)
• With the preference off, or with agent_context absent from stored settings, all three paths leave load_memory at False — test_profile_launch_leaves_memory_off_without_the_preference (unchanged), test_direct_agent_launch_leaves_memory_off_without_the_preference and test_agent_settings_launch_leaves_memory_off_without_the_preference
• An agent that already sets load_memory=True itself is unaffected — test_agent_launch_preserves_context_when_load_memory_already_true
• _resolve_agent_from_profile no longer takes a load_memory parameter — confirmed in the diff (signature, docstring, and body)
• Existing tests in tests/agent_server/test_agent_profile_conv_start.py pass unchanged — confirmed, all 39 passed

Video/Screenshots

No video — this is a backend-only change with no UI. The script above is the functional evidence: it exercises the real settings store and a real MEMORY.md on disk rather than pytest mocks, and its output is included in How to Test section above.

Type

• Bug fix
• Feature
• Refactor
• Breaking change
• Docs / chore

Notes

• User-facing: the global "load persistent memory" preference now applies to every conversation launch (agent, agent_settings, agent_profile_id), not only ones started from a named agent profile.

• _with_load_memory operates on AgentBase generically, so it isn't kind-specific — the existing profile tests already cover the openhands/acp split, and since this fix's logic doesn't branch on agent kind, that split doesn't need re-testing for the new shapes.

• load_skills_from_agent_server() also discards a caller-supplied AgentContext outright — noted in the issue as an optional companion change. Filing that separately rather than bundling it here, since it's an independent code path.

[github-actions]

📁 PR Artifacts Notice

This PR contains a .pr/ directory with temporary PR-specific documents. Because this is a fork PR, the directory will be automatically removed from main immediately after merge.

[Copilot]

Pull request overview

Fixes propagation of the persisted load_memory preference across all conversation launch paths.

Changes:

• Centralizes memory-preference application after agent resolution.
• Removes the obsolete profile-resolution parameter.
• Adds regression tests and real-store verification.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Review notes
tests/agent_server/test_agent_profile_conv_start.py	Adds coverage for direct-agent and agent_settings launches.
openhands-agent-server/openhands/agent_server/conversation_service.py	Nit (2 votes): maintainer evaluation and prompt-impact review are requested.
.pr/verify_load_memory_all_paths.py	Critical (1 vote): verification can overwrite an existing settings.json; isolate or restore the settings store.

Suppressed comments (1)

openhands-agent-server/openhands/agent_server/conversation_service.py:1470

• FileSettingsStore.load() performs synchronous file I/O and full settings validation, but this call now runs on the event loop for every concrete-agent launch (not just profile launches). A slow or large settings file can therefore block unrelated requests while conversations start; move the load to the worker thread, as the surrounding profile resolution already does.

        settings = get_settings_store().load() or PersistedSettings()

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[all-hands-bot]

🚦 CI is currently failing on this PR's latest commit.

Please fix the failing checks before OpenHands reviews it - this is re-checked automatically once you push a new commit. (A maintainer can also request @all-hands-bot as a reviewer to have it reviewed regardless of CI status.)

This is an automated check - no AI was used to generate this comment.

[all-hands-bot]

🚦 CI is currently failing on this PR's latest commit.

Please fix the failing checks before OpenHands reviews it - this is re-checked automatically once you push a new commit. (A maintainer can also request @all-hands-bot as a reviewer to have it reviewed regardless of CI status.)

This is an automated check - no AI was used to generate this comment.