Get started

Get started

Run the demonstration pack. One sequence.

From a checkout you already have: pip install -e ., or a wheel built from that checkout. Then etalon pack init --from demo writes mock_endpoint.py. Start that fixture, then qualify etalon.demo.ticket-routing@0.1.1. That pack is an example. It is not the commercial contact-routing pack. A QUALIFIED result does not grant permission to use the model without human review.

Open the quickstartQuoted steps
View the sample report on this siteDocs indexWhich pack is which

One sequence

A checkout you already have. Install it with pip install -e ., or install a wheel built from that checkout. Then the same pack init. Python 3.11 or newer.

A fixture the init command writes

etalon pack init --from demo writes mock_endpoint.py. You do not need that file to already be in the tree. It is a script, not model weights.

No Coyos account

The demonstration does not need Coyos in the loop. The runner does not send product telemetry.

The demonstration

Init writes the fixture. Then qualify.

Checkout and wheel use the same commands after install. This is the sequence on getting started. etalon --help on the installed version wins on flags.

  1. 1

    Install from that tree

    Python 3.11 or newer, in a checkout you already have. pip install -e . is one install. A wheel built from that same checkout is the other. Either one continues with the next step.

    python3 -m venv .venv
source .venv/bin/activate
pip install -e .
# or, instead of the editable install, a wheel from this checkout:
# python3 -m build
# pip install dist/etalon-0.0.2-*.whl
etalon --version

    etalon --version prints etalon 0.0.2. PyPI etalon==0.0.2 prints the same line and does not write mock_endpoint.py.

  2. 2

    Write the pack and the fixture

    etalon pack init --from demo ./demo-pack writes ./demo-pack and demo-pack/mock_endpoint.py. The fixture is a script, not model weights. It does not have to exist before this command.

    etalon pack init --from demo ./demo-pack
etalon pack validate ./demo-pack
  3. 3

    Start the local fixture

    It listens on 127.0.0.1 only. Leave it running.

    python3 demo-pack/mock_endpoint.py --port 8000
  4. 4

    Qualify the demonstration pack

    Model etalon-demo-baseline is the scripted profile whose completed capture is QUALIFIED. A QUALIFIED result does not grant permission to use the model without human review.

    etalon qualify \
  --endpoint http://127.0.0.1:8000/v1 \
  --pack ./demo-pack \
  --serving-config ./demo-pack/serving.yaml \
  --system-prompt ./demo-pack/system_prompt.txt \
  --model etalon-demo-baseline \
  --operator "Demo Operator" \
  --output ./evidence
  5. 5

    Inspect, verify, and open the report

    Replace <run-id> with the directory name qualify printed. report prints the existing report.html path and does not rewrite the bundle.

    etalon inspect ./evidence/<run-id>
etalon verify ./evidence/<run-id>
etalon report ./evidence/<run-id>

Three different packs

The demonstration pack is not the commercial contact-routing pack.

Example customers can run

Demonstration pack

etalon.demo.ticket-routing@0.1.1 in ./demo-pack, written by etalon pack init --from demo. Synthetic tickets. No judge. This is the sequence above.

Example customers can run

Public contact-routing fixture

coyos.fs.contact-routing@0.3.1 in examples/contact-routing on a runner checkout. Synthetic. TypeSafe judge. Not written by pack init --from demo. Not for sale. Details are on the packs page in the docs.

Commercial pack customers can buy

Commercial contact-routing pack

The pack you can buy. Id coyos.fs.contact-routing, at the version in the delivery, not coyos.fs.contact-routing@0.3.1. Your routing task, your cases, and your acceptance criteria. It is not the Coyos service. Ask for it. A QUALIFIED result does not grant permission to use the model without human review.

Coding agent

Point it at the same page.

The prompt is the same sequence: checkout or wheel, then pack init, which writes the fixture, then qualify, inspect, verify, and report.

Open skill.md
Prompt to paste
Follow this sequence. It is the same sequence as https://docs.etalon.sh/docs/getting-started. Do not invent a second install path. If etalon --help disagrees on flags, the installed help wins.
The demonstration pack is etalon.demo.ticket-routing@0.1.1. It is an example. It is not the commercial contact-routing pack.
The public fixture coyos.fs.contact-routing@0.3.1 in examples/contact-routing is a second example in a runner checkout. It declares a TypeSafe judge. It is not the commercial pack, and this demonstration does not use it. Do not start there unless I ask.
Python 3.11 or newer. Start from a checkout you already have. The fixture does not have to exist before pack init.
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
# or, instead of the editable install, a wheel built from this checkout:
# python3 -m build
# pip install dist/etalon-0.0.2-*.whl
etalon --version
That prints etalon 0.0.2. PyPI etalon==0.0.2 prints the same line and does not write mock_endpoint.py.
Then, from either install:
etalon pack init --from demo ./demo-pack
etalon pack validate ./demo-pack
That command writes ./demo-pack and demo-pack/mock_endpoint.py. The fixture is a script, not model weights.
In one terminal:
python3 demo-pack/mock_endpoint.py --port 8000
In a second terminal, with the virtual environment active:
etalon qualify --endpoint http://127.0.0.1:8000/v1 --pack ./demo-pack --serving-config ./demo-pack/serving.yaml --system-prompt ./demo-pack/system_prompt.txt --model etalon-demo-baseline --operator "Demo Operator" --output ./evidence
Then, with the run directory qualify printed:
etalon inspect ./evidence/<run-id>
etalon verify ./evidence/<run-id>
etalon report ./evidence/<run-id>
A QUALIFIED result does not grant permission to use the model without human review.
Do not send product telemetry. Evaluation requests go only to the endpoint I configure and, where enabled, a judge service.
Or install the skill
# into your agent's skills directory, for example:
mkdir -p .cursor/skills/etalon
curl -fsSL https://www.etalon.sh/skill.md -o .cursor/skills/etalon/SKILL.md

Review the skill before use. The getting-started page remains the procedure.

Then

After the demonstration.

The demonstration does not need us. A commercial pack does.

Run the quickstart on your own. Write to us for the commercial contact-routing pack, or for the service that establishes a process for one other task.

Commercial packCoyos service