Metadata-Version: 2.5
Name: slmeval-cli
Version: 0.1.1
Summary: Benchmark agent workflows across small language models, locally, from your terminal.
Project-URL: Homepage, https://github.com/mohitraj3697/slmeval
Project-URL: Documentation, https://github.com/mohitraj3697/slmeval/tree/main/docs
Project-URL: Changelog, https://github.com/mohitraj3697/slmeval/blob/main/CHANGELOG.md
Project-URL: Issues, https://github.com/mohitraj3697/slmeval/issues
Author: Mohit Raj
License:                               Apache License
                                Version 2.0, January 2004
                             http://www.apache.org/licenses/
        
        TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
        1. Definitions.
        
           "License" shall mean the terms and conditions for use, reproduction,
           and distribution as defined by Sections 1 through 9 of this document.
        
           "Licensor" shall mean the copyright owner or entity authorized by
           the copyright owner that is granting the License.
        
           "Legal Entity" shall mean the union of the acting entity and all
           other entities that control, are controlled by, or are under common
           control with that entity. For the purposes of this definition,
           "control" means (i) the power, direct or indirect, to cause the
           direction or management of such entity, whether by contract or
           otherwise, or (ii) ownership of fifty percent (50%) or more of the
           outstanding shares, or (iii) beneficial ownership of such entity.
        
           "You" (or "Your") shall mean an individual or Legal Entity
           exercising permissions granted by this License.
        
           "Source" form shall mean the preferred form for making modifications,
           including but not limited to software source code, documentation
           source, and configuration files.
        
           "Object" form shall mean any form resulting from mechanical
           transformation or translation of a Source form, including but
           not limited to compiled object code, generated documentation,
           and conversions to other media types.
        
           "Work" shall mean the work of authorship, whether in Source or
           Object form, made available under the License, as indicated by a
           copyright notice that is included in or attached to the work
           (an example is provided in the Appendix below).
        
           "Derivative Works" shall mean any work, whether in Source or Object
           form, that is based on (or derived from) the Work and for which the
           editorial revisions, annotations, elaborations, or other modifications
           represent, as a whole, an original work of authorship. For the purposes
           of this License, Derivative Works shall not include works that remain
           separable from, or merely link (or bind by name) to the interfaces of,
           the Work and Derivative Works thereof.
        
           "Contribution" shall mean any work of authorship, including
           the original version of the Work and any modifications or additions
           to that Work or Derivative Works thereof, that is intentionally
           submitted to Licensor for inclusion in the Work by the copyright owner
           or by an individual or Legal Entity authorized to submit on behalf of
           the copyright owner. For the purposes of this definition, "submitted"
           means any form of electronic, verbal, or written communication sent
           to the Licensor or its representatives, including but not limited to
           communication on electronic mailing lists, source code control systems,
           and issue tracking systems that are managed by, or on behalf of, the
           Licensor for the purpose of discussing and improving the Work, but
           excluding communication that is conspicuously marked or otherwise
           designated in writing by the copyright owner as "Not a Contribution."
        
           "Contributor" shall mean Licensor and any individual or Legal Entity
           on behalf of whom a Contribution has been received by Licensor and
           subsequently incorporated within the Work.
        
        2. Grant of Copyright License. Subject to the terms and conditions of
           this License, each Contributor hereby grants to You a perpetual,
           worldwide, non-exclusive, no-charge, royalty-free, irrevocable
           copyright license to reproduce, prepare Derivative Works of,
           publicly display, publicly perform, sublicense, and distribute the
           Work and such Derivative Works in Source or Object form.
        
        3. Grant of Patent License. Subject to the terms and conditions of
           this License, each Contributor hereby grants to You a perpetual,
           worldwide, non-exclusive, no-charge, royalty-free, irrevocable
           (except as stated in this section) patent license to make, have made,
           use, offer to sell, sell, import, and otherwise transfer the Work,
           where such license applies only to those patent claims licensable
           by such Contributor that are necessarily infringed by their
           Contribution(s) alone or by combination of their Contribution(s)
           with the Work to which such Contribution(s) was submitted. If You
           institute patent litigation against any entity (including a
           cross-claim or counterclaim in a lawsuit) alleging that the Work
           or a Contribution incorporated within the Work constitutes direct
           or contributory patent infringement, then any patent licenses
           granted to You under this License for that Work shall terminate
           as of the date such litigation is filed.
        
        4. Redistribution. You may reproduce and distribute copies of the
           Work or Derivative Works thereof in any medium, with or without
           modifications, and in Source or Object form, provided that You
           meet the following conditions:
        
           (a) You must give any other recipients of the Work or
               Derivative Works a copy of this License; and
        
           (b) You must cause any modified files to carry prominent notices
               stating that You changed the files; and
        
           (c) You must retain, in the Source form of any Derivative Works
               that You distribute, all copyright, patent, trademark, and
               attribution notices from the Source form of the Work,
               excluding those notices that do not pertain to any part of
               the Derivative Works; and
        
           (d) If the Work includes a "NOTICE" text file as part of its
               distribution, then any Derivative Works that You distribute must
               include a readable copy of the attribution notices contained
               within such NOTICE file, excluding those notices that do not
               pertain to any part of the Derivative Works, in at least one
               of the following places: within a NOTICE text file distributed
               as part of the Derivative Works; within the Source form or
               documentation, if provided along with the Derivative Works; or,
               within a display generated by the Derivative Works, if and
               wherever such third-party notices normally appear. The contents
               of the NOTICE file are for informational purposes only and
               do not modify the License. You may add Your own attribution
               notices within Derivative Works that You distribute, alongside
               or as an addendum to the NOTICE text from the Work, provided
               that such additional attribution notices cannot be construed
               as modifying the License.
        
           You may add Your own copyright statement to Your modifications and
           may provide additional or different license terms and conditions
           for use, reproduction, or distribution of Your modifications, or
           for any such Derivative Works as a whole, provided Your use,
           reproduction, and distribution of the Work otherwise complies with
           the conditions stated in this License.
        
        5. Submission of Contributions. Unless You explicitly state otherwise,
           any Contribution intentionally submitted for inclusion in the Work
           by You to the Licensor shall be under the terms and conditions of
           this License, without any additional terms or conditions.
           Notwithstanding the above, nothing herein shall supersede or modify
           the terms of any separate license agreement you may have executed
           with Licensor regarding such Contributions.
        
        6. Trademarks. This License does not grant permission to use the trade
           names, trademarks, service marks, or product names of the Licensor,
           except as required for reasonable and customary use in describing the
           origin of the Work and reproducing the content of the NOTICE file.
        
        7. Disclaimer of Warranty. Unless required by applicable law or
           agreed to in writing, Licensor provides the Work (and each
           Contributor provides its Contributions) on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
           implied, including, without limitation, any warranties or conditions
           of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
           PARTICULAR PURPOSE. You are solely responsible for determining the
           appropriateness of using or redistributing the Work and assume any
           risks associated with Your exercise of permissions under this License.
        
        8. Limitation of Liability. In no event and under no legal theory,
           whether in tort (including negligence), contract, or otherwise,
           unless required by applicable law (such as deliberate and grossly
           negligent acts) or agreed to in writing, shall any Contributor be
           liable to You for damages, including any direct, indirect, special,
           incidental, or consequential damages of any character arising as a
           result of this License or out of the use or inability to use the
           Work (including but not limited to damages for loss of goodwill,
           work stoppage, computer failure or malfunction, or any and all
           other commercial damages or losses), even if such Contributor
           has been advised of the possibility of such damages.
        
        9. Accepting Warranty or Additional Liability. While redistributing
           the Work or Derivative Works thereof, You may choose to offer,
           and charge a fee for, acceptance of support, warranty, indemnity,
           or other liability obligations and/or rights consistent with this
           License. However, in accepting such obligations, You may act only
           on Your own behalf and on Your sole responsibility, not on behalf
           of any other Contributor, and only if You agree to indemnify,
           defend, and hold each Contributor harmless for any liability
           incurred by, or claims asserted against, such Contributor by reason
           of your accepting any such warranty or additional liability.
        
        END OF TERMS AND CONDITIONS
        
        APPENDIX: How to apply the Apache License to your work.
        
           To apply the Apache License to your work, attach the following
           boilerplate notice, with the fields enclosed by brackets "[]"
           replaced with your own identifying information. (Don't include
           the brackets!)  The text should be enclosed in the appropriate
           comment syntax for the file format. We also recommend that a
           file or class name and description of purpose be included on the
           same "printed page" as the copyright notice for easier
           identification within third-party archives.
        
        Copyright 2026 slmeval contributors
        
        Licensed under the Apache License, Version 2.0 (the "License");
        you may not use this file except in compliance with the License.
        You may obtain a copy of the License at
        
            http://www.apache.org/licenses/LICENSE-2.0
        
        Unless required by applicable law or agreed to in writing, software
        distributed under the License is distributed on an "AS IS" BASIS,
        WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
        See the License for the specific language governing permissions and
        limitations under the License.
License-File: LICENSE
Keywords: agents,benchmark,ci,evaluation,langgraph,llm,ollama,slm
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Software Development :: Testing
Classifier: Typing :: Typed
Requires-Python: >=3.11
Requires-Dist: httpx>=0.27
Requires-Dist: pydantic>=2.7
Requires-Dist: rich>=13.7
Requires-Dist: typer>=0.12
Provides-Extra: anthropic
Requires-Dist: anthropic>=0.34; extra == 'anthropic'
Provides-Extra: dev
Requires-Dist: mypy>=1.11; extra == 'dev'
Requires-Dist: pytest-asyncio>=0.23; extra == 'dev'
Requires-Dist: pytest>=8.0; extra == 'dev'
Requires-Dist: pyyaml>=6.0; extra == 'dev'
Requires-Dist: ruff>=0.6; extra == 'dev'
Provides-Extra: langgraph
Requires-Dist: langchain-openai>=0.2; extra == 'langgraph'
Requires-Dist: langgraph>=0.2; extra == 'langgraph'
Description-Content-Type: text/markdown

# slmeval

[![CI](https://github.com/mohitraj3697/slmeval/actions/workflows/ci.yml/badge.svg)](https://github.com/mohitraj3697/slmeval/actions/workflows/ci.yml)
[![PyPI](https://img.shields.io/pypi/v/slmeval-cli)](https://pypi.org/project/slmeval-cli/)
[![Python](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13-blue)](https://github.com/mohitraj3697/slmeval/blob/main/pyproject.toml)
[![License](https://img.shields.io/badge/license-Apache%202.0-blue)](https://github.com/mohitraj3697/slmeval/blob/main/LICENSE)

Test whether a small model can take over steps of your agent, without calling
your real tools again and without a GPU in CI.

slmeval records your agent's tool calls once, then replays those responses while
each candidate model runs your test cases. You get a scorecard per model and per
agent step, and a command that fails the build when a model falls below your
thresholds.

![Terminal output of slmeval report: a warning about a forbidden tool call, a table comparing four models, and a table showing which agent step each model failed at](https://github.com/mohitraj3697/slmeval/raw/main/docs/assets/scorecard.svg)

Output from the bundled demo, shortened. **The four models are scripted stand-ins,
not real SLMs**: each one is written to fail in a known way, to show what the tool
reports. These numbers say nothing about any real model.

## Why

Parts of most agents, such as routing, extraction and formatting, may not need a
frontier model. Finding out is awkward: every test run calls your real tools, which
costs money and can have side effects, and the tool results change between runs.

With slmeval, the tool responses stay fixed, so the model is the only thing that
changes. Runs repeat exactly on the tool side and need no network for tools. You can
also swap the model on a single step and leave the rest of the agent alone.

## Try the demo

```bash
git clone https://github.com/mohitraj3697/slmeval
cd slmeval/examples/support_agent
pip install slmeval-cli
slmeval record
slmeval run
```

This takes under a minute and needs no API key or GPU. `record` runs the agent's
tools against an in-memory stand-in for an order system. `run` replays them against
four models.

The package is `slmeval-cli` on PyPI. The command and the Python import are both
`slmeval`. `slm-eval` on PyPI is a different project.

## What the demo shows

The agent is a small customer support bot with four steps and five tools. The
results below come from [Teardown #1](https://github.com/mohitraj3697/slmeval/tree/main/teardown),
which publishes every output along with the config that produced it.

| Model | What it is scripted to do | What slmeval reports |
|---|---|---|
| `baseline` | Answers correctly | The reference for the others |
| `fenced-json` | Answers correctly, but wraps its JSON in prose and a code fence | Passes every case, with 0% schema-valid output on the first try |
| `sloppy-router` | Drops the `ORD-` prefix from order ids | A failed tool lookup, with a diff of the arguments |
| `refund-happy` | Claims the customer confirmed a refund they only asked for | A forbidden tool call, which fails the gate at any pass rate |

| Candidate | pass^3 | 95% CI | runs | Schema | Tool F1 | Steps |
|---|---|---|---|---|---|---|
| **baseline** | 100.0% | [51%, 100%] | 12/12 | 100.0% | 1.00 | 4.8 |
| sloppy-router | 33.3% | [6%, 79%] | 3/9 | 100.0% | 0.40 | 1.8 |
| fenced-json | 100.0% | [51%, 100%] | 12/12 | 0.0% | 1.00 | 4.8 |
| refund-happy | 75.0% | [30%, 95%] | 9/12 | 100.0% | 0.96 | 4.8 |

`pass^3` is the share of cases that passed all three repeats. The demo uses only four
cases, so the intervals are wide. A real comparison needs far more.

### Pass rates come with intervals

![Chart of pass rate and 95 percent interval for each model. baseline and fenced-json reach 100 percent with a lower bound of 51 percent, which clears a gate floor of 50 percent. sloppy-router and refund-happy fall below it.](https://github.com/mohitraj3697/slmeval/raw/main/docs/assets/pass-rates.svg)

Passing four cases out of four still only gives a lower bound of 51%. `slmeval gate`
checks the lower bound by default, so a small sample can't pass on its point estimate
alone.

### Failures point to a step

![Grid of models against agent steps. sloppy-router failed 6 times at order_lookup, refund-happy failed 3 times at policy_check, and the other steps passed or made no model call.](https://github.com/mohitraj3697/slmeval/raw/main/docs/assets/where-models-failed.svg)

When a model asks for something that was never recorded, slmeval finds the closest
recorded call and shows what differs:

```text
✗ status_shipped · sloppy-router · node=order_lookup · CASSETTE_MISS
  tool: fetch_order
  called with: {"order_id": "9921", "request_id": "req-9921"}
  nearest recorded: {"order_id": "ORD-9921", "request_id": "req-ORD-9921"}
  diff: order_id: "ORD-9921" → "9921"   (dropped prefix)
        request_id: "req-ORD-9921" → "req-9921"   (different value)
  verdict: hallucinated_parameter
```

If the model took a reasonable path that the baseline just never recorded, slmeval
reports `unrecorded_valid_path` instead and leaves that run out of the pass rate.
That is why `sloppy-router` shows 3/9 runs and not 3/12.

### Fault injection finds agent bugs too

![Chart comparing pass rates with no faults and with 30 percent of tool calls faulted. baseline and fenced-json drop from 100 to 50 percent, refund-happy from 75 to 50 percent, and sloppy-router stays at 33 percent.](https://github.com/mohitraj3697/slmeval/raw/main/docs/assets/fault-injection.svg)

Re-running with injected tool failures found a bug in our own demo agent: it crashes
with `KeyError: 'carrier'` when a tool returns an empty response. No model could
recover from that, including the baseline. We left the bug in the demo because it
shows what fault injection is for.

## Use it on your agent

Install it and create a starter config:

```bash
pip install slmeval-cli
slmeval init
slmeval doctor
```

Then mark the model calls and tools in your agent. Both helpers do nothing outside a
benchmark, so they are safe to leave in production code:

```python
from slmeval import chat_model, node, sandbox_tool

@sandbox_tool(match_on=["order_id"], ignore_args=["request_id"])
async def fetch_order(order_id: str, request_id: str) -> dict:
    ...  # recorded once, then replayed

async def run_case(case_input: dict) -> dict:
    with node("intent_router"):  # names this step in the report
        llm = chat_model(provider="openai", model="gpt-4o-mini")  # used outside a benchmark
        completion = await llm.complete(messages, schema=Routing)
    ...
```

Point `eval.toml` at that function and list the models to try. `target_nodes` swaps
the model on the named steps only:

```toml
[target]
entrypoint = "my_agent:run_case"

[baseline]
provider = "openai"
model = "gpt-4o-mini"

[[candidates]]
id = "gemma3-router"
provider = "ollama"
model = "gemma3:4b"
target_nodes = ["intent_router"]
```

```bash
slmeval record
slmeval run
slmeval show run
```

For LangGraph, `slmeval.instrument_langgraph(builder)` replaces the `node()` blocks.
The [getting started guide](https://github.com/mohitraj3697/slmeval/blob/main/docs/getting-started.md)
covers the rest, including how to write test cases.

## Use it in CI

```bash
slmeval run --tool-mode replay
slmeval gate --min-pass-k 0.90 --max-cost-ratio 0.40
```

Tool responses come from a local SQLite file, so CI needs no tool credentials. If
your candidates are hosted or local models, the runner still needs to reach them.
Use the `fake` provider for a check with no model server at all.

| Exit code | Meaning |
|---|---|
| 0 | Every threshold was met |
| 1 | A threshold was missed, or a forbidden tool was called |
| 2 | Bad invocation, such as no threshold set |
| 3 | A model server was unreachable, so nothing was measured |
| 4 | Not enough recorded responses to judge the run |

There is also a [GitHub Action](https://github.com/mohitraj3697/slmeval/tree/main/examples/github-action)
that writes the scorecard to the job summary and keeps one comment updated on the pull
request:

```yaml
- uses: actions/checkout@v4
- uses: mohitraj3697/slmeval/examples/github-action@main
  with:
    working-directory: evals
    cassettes: evals/cassettes.sqlite
```

## How it works

![Diagram. Once: your agent and its real tools go through slmeval record into cassettes.sqlite. Every run: candidates go through slmeval run, which reads the cassettes, then checks, a report, and slmeval gate.](https://github.com/mohitraj3697/slmeval/raw/main/docs/assets/how-it-works.svg)

- **Tools** are recorded once and served from SQLite after that. Return types such as
  `datetime` and `Decimal` come back as the same types.
- **Checks** run in tiers: assertions from your dataset, then tool call and path
  comparisons against the baseline, then pass rates across repeats. An LLM judge exists
  but isn't connected to `slmeval run` yet.
- **Results** are stored per run, so you can re-render a report or inspect a single
  execution later without loading your agent.

## Status

Alpha. Version 0.1.1 works end to end, and the exit codes and JSON report schema are
treated as stable.

- **Providers:** OpenAI, Ollama, anything that speaks `/v1/chat/completions`, and a
  scripted `fake` provider for offline runs.
- **Not built yet:** `slmeval compare`, `slmeval cassette`, a native Anthropic provider
  and time-to-first-token measurement.
- **Results for real models are limited so far.** Teardown #1 uses scripted models.
  [Teardown #2](https://github.com/mohitraj3697/slmeval/tree/main/teardown-02) runs three
  local models on one laptop, with a local model as the baseline. If you run your own,
  [share the result](https://github.com/mohitraj3697/slmeval/issues/new?template=benchmark_result.yml).

slmeval sends no telemetry. Its only network traffic is to the model endpoints you
configure, and `eval.toml` holds the name of an API key's environment variable, never
the key itself.

## Documentation

| | |
|---|---|
| [Getting started](https://github.com/mohitraj3697/slmeval/blob/main/docs/getting-started.md) | Connect your agent and write test cases |
| [Reading results](https://github.com/mohitraj3697/slmeval/blob/main/docs/reading-results.md) | What each number means, and how to drill into a failure |
| [Configuration](https://github.com/mohitraj3697/slmeval/blob/main/docs/configuration.md) | Providers, structured output, the `fake` provider and fault injection |
| [CI](https://github.com/mohitraj3697/slmeval/blob/main/docs/ci.md) | Gate thresholds, baselines and the GitHub Action |
| [Architecture](https://github.com/mohitraj3697/slmeval/blob/main/docs/architecture.md) | How the pieces fit, what is stored, and the LLM judge |
| [Teardown #1](https://github.com/mohitraj3697/slmeval/tree/main/teardown) | The demo's full results and their limitations |

## Contributing

Bug reports and pull requests are welcome. See
[CONTRIBUTING.md](https://github.com/mohitraj3697/slmeval/blob/main/CONTRIBUTING.md). To
report a security issue, follow
[SECURITY.md](https://github.com/mohitraj3697/slmeval/blob/main/SECURITY.md).

## License

[Apache 2.0](https://github.com/mohitraj3697/slmeval/blob/main/LICENSE)
