Repository navigation
Conversation
Wrap Context Retrieval on the facade as `context(query, *, user_id, top_k, include_profile, filters, app_id, project_id)`, the last segment of its route like the other memory methods. user_id is required: the loop returns episodes for user owners only. The body carries none of the fields the endpoint rejects with 422 (min_score, radius, enable_llm_rerank, with_readable_episode). Quickstart and README list the method. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
gloryfromca
left a comment
There was a problem hiding this comment.
Blocking: the generated context model and endpoint artifacts must be included so the package imports again.
I found one blocker, marked inline. In a fresh virtual environment with the declared runtime and test dependencies, pytest -q stops during collection with 1 error because EverOS is no longer exported; both Python CI jobs reproduce the same import failure.
I covered the full diff, surrounding facade/model/API callers, recent facade and generated-SDK history, public backward compatibility, test strength, and the facade/generated-client architecture. I found no AGENTS.md, CLAUDE.md, or CONTEXT*.md repository guidance to apply. The new tests were not weakened, but they cannot collect against the artifacts in this revision. The facade signature, default preservation, scoping, and docs otherwise look consistent with nearby memory methods.
| AddOperation, | ||
| Content, | ||
| ContentItem, | ||
| ContextInput, |
There was a problem hiding this comment.
Blocking: include the generated context endpoint and model before this facade lands.
ContextInput is absent from the package, and MemoryApi does not define retrieve_context. Importing everos_cloud.client therefore raises ImportError; because the top-level __init__.py swallows that error, from everos_cloud import EverOS fails too, breaking the entire existing facade rather than only context(). The clean-environment test run stops during collection on this import, and both CI versions reproduce it. Please add or regenerate the 1.3.0 model, API, exports, and contract artifacts alongside this wrapper.
Adds
client.context()for Context Retrieval, the multi-round LLM retrieval endpoint added in contract 2.0.0 at enginerelease-20261008_v15.context(query, *, user_id, top_k=None, include_profile=None, filters=None, app_id=None, project_id=None)→SearchData, same shape assearch().episodes[].citedmarks what the model's answer rested on.user_idis required: the loop returns episodes for user owners only, soagent_idis left out./contextrejects with 422 (min_score,radius,enable_llm_rerank,with_readable_episode). A test covers this.retrieve_context→contextin the AttributeError hint table. Quickstart and README list the method.Draft until the generated 1.3.0-rc1 lands on
v1(factory MR !14).ContextInputandMemoryApi.retrieve_contextcome from that generation, so CI here fails until then. Tested locally against that generated tree: 60 passed.🤖 Generated with Claude Code