fix(python-sdks): stop mishandling SDK model objects in memory dedup and formatting

[abhay-codes07]

Fixes #1266

What

The pipecat, cartesia, and agent-framework packages all feed search results from the supermemory SDK — pydantic models with attribute access and snake_case fields — into dedup/format helpers written for plain dicts. The failure mode differs per package, but in each one the core feature is broken on its primary path:

• pipecat: deduplicate_memories calls r.get("memory", "") on a Result model → AttributeError → caught by the generic frame handler, frame forwarded unchanged. Memories are never injected for any user whose profile lookup returns search results; it only "works" for brand-new users with empty results.
• cartesia: identical crash, swallowed one level up — memory_context silently comes back None and the process method then strips the previous memory block from the system prompt. Persistent recall is dead for users with stored memories.
• agent-framework: extract_memory_text handles only dict/str, so model objects fall through to return None and every search-result memory is silently dropped (no crash, just data loss).

The OpenAI package shares the same helper shape but works, because it fetches profile data over raw aiohttp and receives dicts — that asymmetry is what hid this.

Fix

• pipecat/cartesia: a new extract_search_result_fields helper reads memory/updatedAt off dicts (camelCase keys) or SDK models (snake_case attributes, tolerant of datetime values), and both deduplicate_memories and format_memories_to_text go through it. The formatter also stops printing raw object reprs for non-dict, non-str items.
• agent-framework: extract_memory_text gains a model-object branch (getattr(item, "memory", None)), keeping the existing trim/empty filtering.

Dict-shaped inputs behave exactly as before in all three packages.

Testing

• Verified against the real SDK model: built supermemory.types.search_memories_response.Result from an API-shaped payload and ran it through dedup → extract → format. Before: AttributeError: 'Result' object has no attribute 'get'. After: renders - [1 Jan] User prefers async.
• New regression tests: tests/test_search_result_objects.py in pipecat and cartesia (dedup without crashing, priority vs profile entries, no-repr formatting, dict compatibility, datetime updated_at, missing fields), plus two model-object cases in agent-framework's tests/test_utils.py.
• Suites: pipecat 7 passed, cartesia 7 passed, agent-framework 56 passed (full suite, in a venv with agent-framework-core==1.0.0rc3 — as noted in Python SDK packages crash or silently drop memories: dedup helpers treat pydantic search results as dicts (pipecat, cartesia, agent-framework) #1266, newer releases dropped BaseContextProvider and break the package import independently of this change).

cc @MaheshtheDev

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.