langchain-finic 0.1.0a8 — Alpha

Highlights

• Preserve configured, bound and configurable decision models when native tool
  arguments omit model; retain explicit request-model overrides.
• Document typed decision-tool argument conversion and credential-free application
  testing, and clarify the structured return type of custom scoring calls.

Installation

pip install langchain-finic==0.1.0a8
# Optional agent middleware:
pip install 'langchain-finic[middleware]==0.1.0a8'

Compatibility

• Python >=3.11.
• httpx>=0.28.1,<1
• langchain-core>=1,<2
• langsmith>=0.3,<1
• pydantic>=2.12,<3
• The middleware extra additionally requires langchain>=1,<2, langgraph>=1,<2.

Checksums and provenance

The attached wheel and source distribution have SHA-256 hashes in SHA256SUMS.
Put the wheel and source distribution in a dist/ directory beside SHA256SUMS, then run:

sha256sum --check SHA256SUMS

On macOS, use shasum -a 256 --check SHA256SUMS. Registry files are published
with attestations through GitHub Actions Trusted Publishing.

Documentation · PyPI ·
Source at this tag

langchain-finic 0.1.0a7 — Alpha

Highlights

• Add a versioned synthetic System 1 evaluation corpus and bounded opt-in runner,
  with frozen held-out configuration, failure-inclusive metrics and explicit limitations.

• Type public middleware and Runnable interfaces, including dictionary inputs and exception-bearing batch results; check consumer type assertions on current and minimum dependencies.

• Include registered tool semantics in screening and add bounded context-builder hooks for routing and screening.

• Explain calibration warm-up, question versioning, application deadlines and shared connection ownership; add an offline-tested async routing example.

• Require explicit scoring-tool opt-in for schema mutation/bypass; direct scoring flags remain available.
  Existing agents that need those flags must construct tools with allow_schema_update=True
  and/or allow_schema_bypass=True.

• Preserve scoring-tool names and JSON output through bindings and configurable fields.

• Apply scorer concurrency defaults to completion-order batches.

• Record sanitized daily live-test evidence and add installed-package application validation.

Installation

pip install langchain-finic==0.1.0a7
# Optional agent middleware:
pip install 'langchain-finic[middleware]==0.1.0a7'

Compatibility

• Python >=3.11.
• httpx>=0.28.1,<1
• langchain-core>=1,<2
• langsmith>=0.3,<1
• pydantic>=2.12,<3
• The middleware extra additionally requires langchain>=1,<2, langgraph>=1,<2.

Checksums and provenance

The attached wheel and source distribution have SHA-256 hashes in SHA256SUMS.
Put the wheel and source distribution in a dist/ directory beside SHA256SUMS, then run:

sha256sum --check SHA256SUMS

On macOS, use shasum -a 256 --check SHA256SUMS. Registry files are published
with attestations through GitHub Actions Trusted Publishing.

Documentation · PyPI ·
Source at this tag

langchain-finic 0.1.0a6 — Alpha

Highlights

• Add typed sync/async clients for custom scoring, model creation, model listing,
  calibration, and job status, covering all six documented Finic REST endpoints.
• Add FinicScorer, explicitly selected lifecycle tools through FinicToolkit,
  and checkpointed LangGraph composition examples.
• Use single-attempt scoring/creation/calibration writes and explicit bounded job
  polling; retain caller-assigned score IDs for reconciliation.
• Align decisions with documented data/context inputs, base-rate boundaries,
  and structured provider errors. Existing state inputs remain supported;
  serialized requests now use canonical data.
• Preserve legacy state assignment/copy updates and configured or bound decision
  tool conversion, including minimum supported LangChain dependencies.
• Accept documented text outcome labels in JSON and CSV without rewriting inputs.
• Expand opt-in live decision coverage to all question types, bounded batches,
  LangGraph tools, middleware, and data/context inputs (15 checks / 18 requests).
• Add isolated synthetic model lifecycle validation, retain pending score IDs
  before writes, and keep validation active under optimized Python.
• Strengthen offline failure, cancellation, ownership, and partial-batch coverage;
  document compatibility, maintenance, and release-promotion policies.
• Retry exact-version PyPI installation verification with fresh index metadata
  and a finite budget, preserving immutable artifacts and registry hash checks.

The package remains alpha. Model creation and calibration are explicit remote
writes; the separate lifecycle test does not run in ordinary CI. Test coverage
validates API and framework behavior, not fraud-detection accuracy.

Installation

pip install langchain-finic==0.1.0a6
# Optional agent middleware:
pip install 'langchain-finic[middleware]==0.1.0a6'

Compatibility

• Python >=3.11.
• httpx>=0.28.1,<1
• langchain-core>=1,<2
• langsmith>=0.3,<1
• pydantic>=2.12,<3
• The middleware extra additionally requires langchain>=1,<2, langgraph>=1,<2.

Checksums and provenance

The attached wheel and source distribution have SHA-256 hashes in SHA256SUMS.
Put the wheel and source distribution in a dist/ directory beside SHA256SUMS, then run:

sha256sum --check SHA256SUMS

On macOS, use shasum -a 256 --check SHA256SUMS. Registry files are published
with attestations through GitHub Actions Trusted Publishing.

Documentation · PyPI ·
Source at this tag

langchain-finic 0.1.0a5 — Alpha

Highlights

• Add opt-in live LangChain standard tool integration tests alongside sync/async
  classifier smoke tests.
• Clarify independent integration ownership and improve README navigation and examples.
• Generate version-specific release notes and detect prerelease versions automatically.
• Update GitHub Actions for uv setup, artifact upload/download, and Pages configuration/deployment.

Installation

pip install langchain-finic==0.1.0a5
# Optional agent middleware:
pip install 'langchain-finic[middleware]==0.1.0a5'

Compatibility

• Python >=3.11.
• httpx>=0.28.1,<1
• langchain-core>=1,<2
• langsmith>=0.3,<1
• pydantic>=2.12,<3
• The middleware extra additionally requires langchain>=1,<2, langgraph>=1,<2.

Checksums and provenance

The attached wheel and source distribution have SHA-256 hashes in SHA256SUMS.
Put the wheel and source distribution in a dist/ directory beside SHA256SUMS, then run:

sha256sum --check SHA256SUMS

On macOS, use shasum -a 256 --check SHA256SUMS. Registry files are published
with attestations through GitHub Actions Trusted Publishing.

Documentation · PyPI ·
Source at this tag

langchain-finic 0.1.0a4 — Alpha

Highlights

Initial public alpha. See GitHub Releases
for publication dates and distribution checksums.

• Typed Noul, Choice, MultiChoice, and Score decisions with response validation.
• LangChain Runnable composition, tools, sync/async calls, batching, and tracing metadata.
• Optional model-routing and tool-screening middleware for LangChain agents.
• Configurable model name, transport injection, bounded retries, and client lifecycle management.
• MIT license, typed distributions, usage documentation, and API reference.

Configuration uses FINIC_* environment variables. Earlier unpublished prototypes
used GRAYSON_* aliases; those aliases are not supported by this release.

Installation

pip install langchain-finic==0.1.0a4
# Optional agent middleware:
pip install 'langchain-finic[middleware]==0.1.0a4'

Compatibility

• Python >=3.11.
• httpx>=0.28.1,<1
• langchain-core>=1,<2
• langsmith>=0.3,<1
• pydantic>=2.12,<3
• The middleware extra additionally requires langchain>=1,<2, langgraph>=1,<2.

Checksums and provenance

The attached wheel and source distribution have SHA-256 hashes in SHA256SUMS.
Put the wheel and source distribution in a dist/ directory beside SHA256SUMS, then run:

sha256sum --check SHA256SUMS

On macOS, use shasum -a 256 --check SHA256SUMS. Registry files are published
with attestations through GitHub Actions Trusted Publishing.

Documentation · PyPI ·
Source at this tag