jharness-models
OpenAI Chat (Chat Completions API), OpenAI Responses, and Anthropic Messages adapters, plus provider-neutral model composition for the JHarness kernel.
uv add jharness-models
from jharness.models.openai import OpenAIResponsesModel
| Adapter | Runtime tools | Provider-hosted tools | Ordered output |
|---|---|---|---|
| OpenAI Chat | Function calls | None | Content and calls are normalized into ModelResponse.output |
| Anthropic Messages | Client tool_use |
Official web-search preset | Native block order is retained |
| OpenAI Responses | Function and custom calls | Official web-search and image-generation presets | Native Responses item order is retained |
Model modalities describe what the model itself understands or produces. Tool
ownership is separate: RuntimeToolCall is executed by the JHarness runtime, while a
ProviderToolCall records work already executed by the supplier. Both remain
interleaved with ContentPart values in ordered output.
Each protocol profile contains the exact immutable ModelCapabilities returned by
its model client. The default Responses and Messages profile classes remain
provider-tool neutral. The official openai_responses_profile() and
anthropic_messages_profile() factories install their hosted-tool identities and
capabilities, but do not add a tool to any request. Hosted-tool mapping is a closed,
protocol-owned union; profiles cannot inject custom wire codecs. The host must still
pass an explicit ProviderToolSpec factory result to Runtime, and selecting an official profile is
the host's confirmation that the chosen endpoint and model support its advertised
capabilities. Tool selection is declared as a set of supported types rather than a
coarse boolean. Supplier factories only compose protocol capabilities and wire
policies; each adapter emits only its documented protocol wire.
Capabilities are host assertions rather than live endpoint discovery. JHarness blocks
requests outside the selected profile, but an overstated profile may still reach a
provider that rejects, ignores, or degrades the claimed feature. Use an exact,
model-appropriate profile. Image MIME ArtifactRef values count as image inputs:
Responses emits input_image, Anthropic Messages emits an image file source, and Chat
uses the standard nested file content part and therefore also requires file input.
from jharness.kernel import Runtime
from jharness.models.openai import (
OpenAIResponsesModel,
openai_responses_profile,
openai_responses_web_search,
)
model = OpenAIResponsesModel(..., profile=openai_responses_profile())
runtime = Runtime(
model=model,
provider_tools=(openai_responses_web_search(),),
)
OpenAI Responses sends store=false by default and requests encrypted reasoning
history; assistant and reasoning history is replayed only from complete native Responses
output items. Hosted image generation additionally requires a host-owned
OpenAIResponsesArtifactStore; generated base64 is persisted externally and durable history
contains only integrity-bearing ArtifactRef values. Stores must be durable,
idempotent, safe for provider-controlled call ids, available during run recovery, and
responsible for retention and garbage collection of uncommitted saves.
Retry and fallback compose directly around model values:
from jharness.models.decorators import FallbackModel, RetryingModel
model = FallbackModel(
RetryingModel(primary_model, max_attempts=3),
RetryingModel(backup_model, max_attempts=2),
)
Installing this distribution installs the exact matching jharness-kernel version.
Provider configuration and composition details are in the
model adapter guide.
Metadata
Release files for jharness-models 0.9.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| jharness_models-0.9.0.tar.gz | 73.6 kB | With provenance | Details for jharness_models-0.9.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| jharness_models-0.9.0-py3-none-any.whl | Python 3 | none | any | Details for jharness_models-0.9.0-py3-none-any.whl |
Total release size: 167.4 kB
Release files / jharness_models-0.9.0.tar.gz
| Download URL | jharness_models-0.9.0.tar.gz |
|---|---|
| Size | 73.6 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
d99512a44beeed389a7b9f9e30436b094dbbd3f127251e31c345a3213a93ca10
|
|
BLAKE2b-256 checksum How to use checksums |
5812d8b9702db14abae018719b0afebb456f8b8147eb86f65a38082ab690818a
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 30, 2026.
Transparency logRelease files / jharness_models-0.9.0-py3-none-any.whl
| Download URL | jharness_models-0.9.0-py3-none-any.whl |
|---|---|
| Size | 93.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
5cce9798f16397867ad9a0fdd98a3517ebf0f301414bd39078fcaf12232fa187
|
|
BLAKE2b-256 checksum How to use checksums |
9cec0b7e846e1a152218fee15cf821a66ee0166c9e751eacdd455201e4d41823
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 30, 2026.
Transparency logSecurity for jharness-models 0.9.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for jharness_models-0.9.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 30, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.