release: 0.38.0

[stainless-app]

Automated Release PR

0.38.0 (2026-05-01)

Full Changelog: v0.37.0...v0.38.0

Features

• api: api update (92fd43a)
• api: api update (7a8e495)
• api: manual updates (23c9221)
• support setting headers via env (eba0398)

Bug Fixes

• use correct field name format for multipart file arrays (0f569cc)

Performance Improvements

• client: optimize file structure copying in multipart requests (c28640c)

Chores

• internal: more robust bootstrap script (d4a39dc)
• internal: reformat pyproject.toml (19a83c3)
• tests: bump steady to v0.22.1 (7d67d91)

---

This pull request is managed by Stainless's GitHub App.

The semver version number is based on included commit messages. Alternatively, you can manually set the version number in the title of this pull request.

For a better experience, it is recommended to use either rebase-merge or squash-merge when merging this pull request.

🔗 Stainless website
📚 Read the docs
🙋 Reach out for help or questions

[stainless-app]

🧪 Testing

To try out this version of the SDK:

pip install 'https://pkg.stainless.com/s/hyperspell-python/19a83c32f936735d997727bda0bb0ba523f8086c/hyperspell-0.37.0-py3-none-any.whl'

Expires at: Sun, 31 May 2026 05:04:53 GMT
Updated at: Fri, 01 May 2026 05:04:53 GMT

[canaries-inc]

🐤 Canary Summary

This PR is a Python SDK release with no UI/UX changes affecting end users:

• Version bump to 0.38.0 across all configuration files
• Added date parameter to memories API for backend ranking/filtering
• Updated API type definitions and test coverage
• No user-facing UI components or changes present

---

---

View PR tests on Canary

[canaries-inc]

🐤 Canary Proposed Tests

No testable user journeys found for this PR.

[entelligence-ai-pr-reviews]

---

Confidence Score: 5/5 - Safe to Merge

Safe to merge — this PR is a release bump to version 0.38.0 and the automated review found no issues across the reviewed files. No logic changes, security concerns, or correctness problems were identified, and there are no unresolved pre-existing comments to carry forward. The PR appears to be a standard version release with clean diffs.

Key Findings:

• No review comments were generated by the automated analysis, indicating no detectable logic, security, or correctness issues in the changed files.
• Zero critical, significant, or medium-severity issues were flagged by the heuristic analysis, consistent with a version bump or release preparation PR.
• Coverage of 3/8 changed files leaves some files unreviewed, but for a release PR this is a low-risk gap — typically release commits touch changelogs, version files, and lock files which are low-risk by nature.

[entelligence-ai-pr-reviews]

EntelligenceAI PR Summary

Release v0.38.0 of the Hyperspell Python SDK introducing new API parameters, a custom headers env var, a path-selective deep copy utility, and multipart array format control.

• Custom headers via env var: Both Hyperspell and AsyncHyperspell clients now parse HYPERSPELL_CUSTOM_HEADERS (newline-delimited key: value) and merge into default_headers, with explicit headers taking precedence (src/hyperspell/_client.py)
• Memory date parameter: Optional ISO 8601 date field added to MemoryUpdateParams and both sync/async update methods in memories.py
• Search recency_half_life_days: Optional float field added to Options in memory_search_params.py for exponential-decay relevance scoring
• array_format for multipart field names: extract_files / _extract_items now accept an ArrayFormat parameter (brackets, indices, repeat, comma) controlling array segment naming (src/hyperspell/_utils/_utils.py)
• deepcopy_with_paths replaces deepcopy_minimal: New selective shallow-copy utility in src/hyperspell/_files.py; deepcopy_minimal removed from _utils exports and memories.py
• Type aliases relocated: ArrayFormat and NestedFormat moved from _qs.py to _types.py
• Tooling updates: @stdy/cli bumped to 0.22.1; bootstrap script hardened for set -u; MCP badge URLs updated to scoped package name
• Tests: Added/updated tests in test_extract_files.py, test_files.py, and test_memories.py; deleted test_deepcopy.py

---

Confidence Score: 3/5 - Review Recommended

Likely safe but review recommended — this PR introduces useful features including the HYPERSPELL_CUSTOM_HEADERS env var support, an optional date parameter for memory endpoints, and multipart array format control, all of which appear well-structured. However, a pre-existing unresolved concern in src/hyperspell/_utils/_utils.py around the _array_suffix function remains open: when flattened_key is None and the format is 'repeat' or 'comma', the empty-string fallback flattened_key="" is passed to the recursive call instead of properly handling the None case, which could produce malformed query parameter keys. This correctness issue was not introduced by this PR but remains unresolved and affects the multipart array format feature this PR ships.

Key Findings:

• The unresolved comment on _utils.py L140-151 identifies a correctness bug where _array_suffix returning '' for 'repeat' or 'comma' formats causes flattened_key='' (empty string) to be passed recursively instead of None, potentially generating malformed serialized keys like '[0]' or '[key]' with an empty prefix — this is directly related to the new multipart array format control feature.
• The HYPERSPELL_CUSTOM_HEADERS parsing logic (newline-delimited key: value) in _client.py deserves scrutiny for edge cases such as headers with colons in their values, missing separators, or whitespace-only lines, though no specific bug has been filed against it.
• The PR's positive contributions — optional ISO 8601 date field on memory endpoints, path-selective deep copy utility, and multipart format control — are well-scoped SDK additions that follow existing patterns in the codebase.
• Coverage of the review is partial (13 of 22 changed files), meaning some modified files have not been reviewed, leaving some uncertainty about the full correctness of the release.

Files requiring special attention

• src/hyperspell/_utils/_utils.py
• src/hyperspell/_client.py

[entelligence-ai-pr-reviews]

Releases v0.38.0 with a new date field on the memories update endpoint and a targeted deep copy utility replacing the previous full-copy approach in multipart requests.

• Added date: str | datetime | None parameter to sync/async memories.update() and MemoryUpdateParams
• Introduced deepcopy_with_paths in src/hyperspell/_files.py for selective container deep copying along specified key paths
• Removed deepcopy_minimal from src/hyperspell/_utils/_utils.py and its export from src/hyperspell/_utils/__init__.py
• Updated src/hyperspell/resources/memories.py to use deepcopy_with_paths with ['file'] path in both sync and async upload methods
• Added 8 new tests in tests/test_files.py covering deepcopy_with_paths and mutation-safety with extract_files
• Removed tests/test_deepcopy.py along with the deprecated utility
• Version bumped in pyproject.toml, _version.py, and .release-please-manifest.json; OpenAPI spec hash refreshed in .stats.yml

[entelligence-ai-pr-reviews]

Releases hyperspell v0.38.0 with an API enhancement, a deep copy performance improvement, and a test dependency update.

• Added optional date field (Union[str, datetime], ISO 8601) to MemoryUpdateParams and sync/async update methods in memories.py
• Replaced deepcopy_minimal (removed from _utils/_utils.py and _utils/__init__.py) with the new deepcopy_with_paths utility in _files.py, which shallow-copies only containers along specified key paths
• Updated memories.py upload methods to use deepcopy_with_paths imported from _files instead of _utils
• Bumped @stdy/cli from 0.20.2 to 0.22.1 in scripts/mock and scripts/test
• Updated OpenAPI spec URL/hash in .stats.yml
• Removed tests/test_deepcopy.py; added TestDeepcopyWithPaths tests in tests/test_files.py
• Version bumped to 0.38.0 in pyproject.toml, _version.py, and .release-please-manifest.json

[entelligence-ai-pr-reviews]

Release v0.38.0 of the hyperspell Python SDK introducing new API parameters, a custom headers environment variable, and internal utility improvements.

• Custom headers: Both sync and async clients now parse HYPERSPELL_CUSTOM_HEADERS (newline-delimited Key: Value) and merge into default_headers (src/hyperspell/_client.py)
• Array format control: extract_files / _extract_items gain an array_format parameter (brackets, indices, repeat, comma) controlling multipart field name suffixes (src/hyperspell/_utils/_utils.py)
• deepcopy refactor: Replaced deepcopy_minimal with deepcopy_with_paths for path-selective shallow copying; deepcopy_minimal removed from _utils exports (src/hyperspell/_files.py, src/hyperspell/_utils/__init__.py, src/hyperspell/resources/memories.py)
• New API parameters: date (ISO 8601) added to MemoryUpdateParams and memory update methods; recency_half_life_days added to Options in memory_search_params.py
• Type aliases relocated: ArrayFormat and NestedFormat moved from _qs.py to _types.py
• MCP badge URLs: Updated to scoped package name @hyperspell/hyperspell-mcp in README.md
• Scripts: Bootstrap hardened against set -u; @stdy/cli bumped to 0.22.1 in mock/test scripts
• Tests: Added coverage for array_format, deepcopy_with_paths, new API params; removed test_deepcopy.py

[entelligence-ai-pr-reviews]

Correctness: When flattened_key is None and _array_suffix returns "" (for "repeat" or "comma" formats), the new code passes flattened_key="" (empty string) to the recursive call instead of None; the downstream dict-handling branch checks if flattened_key is None to decide whether to assign key or append [key], so an empty string takes the else branch and produces a malformed key like "[fieldname]" instead of "fieldname".

🤖 AI Agent Prompt for Cursor/Windsurf

📋 Copy this prompt to your AI coding assistant (Cursor, Windsurf, etc.) to get help fixing this issue

In `src/hyperspell/_utils/_utils.py`, lines 142–144, the new expression `(flattened_key if flattened_key is not None else "") + _array_suffix(array_format, array_index)` converts a `None` flattened_key into an empty string `""` when `_array_suffix` returns `""` (e.g., for `"repeat"` or `"comma"` array formats). The recursive `_extract_items` call then sees `flattened_key=""`, which is not `None`, so the dict branch appends `[key]` instead of just using `key`, producing a leading-bracket malformed key. Fix by preserving `None` when the suffix is empty: use `(flattened_key + suffix) if flattened_key is not None else (suffix if suffix else None)` where `suffix = _array_suffix(array_format, array_index)`.