Plural

Unified LLM routing with first-class traces, environments, and benchmarks.

Plural is for builders putting AI into products. Start with an OpenRouter-style multi-provider router. Keep going with the thing OpenRouter does not give you: a single Trace object shared by production traffic and RL-style environments — so you can understand prompts, build datasets, run benchmarks, and eventually autoroute to the best model for your data.

flowchart LR
  App[Your app] --> Client[Plural]
  Env[Environment] --> Client
  Client --> Trace[Trace]
  Trace --> TraceDataset[TraceDataset]
  TraceDataset --> Training[Training / export]
  Env --> Task[Task revisions]
  Verifier[Verifier revisions] --> Task
  Task --> Bench[BenchmarkDefinition]
  Agent[AgentDefinition] --> Job[Job]
  Bench --> Job

Install

pip install plural
# optional OpenTelemetry exporter
pip install "plural[otel]"
# optional OS keyring / Daytona sandbox provider
pip install "plural[keyring]" "plural[daytona]"

Quickstart

Set PLURAL_API_KEY, then use the client like the OpenAI SDK:

export PLURAL_API_KEY=plural-...

from plural import Client, Message

client = Client()
assert client.is_authenticated()
response = client.chat(
    model="openai/gpt-4o-mini",
    messages=[Message(role="user", content="Hello from plural")],
    models=["anthropic/claude-sonnet-4"],  # optional fallbacks
)
print(response.text)
client.close()
# Traces → .plural/traces.jsonl

Create or update hosted records with client.create(x) and client.update(x). A project API key already knows the project; an account key needs project= or PLURAL_PROJECT.

Optional BYOK (pass your own upstream keys explicitly):

import os
from plural import Client

client = Client(providers={"openai": os.environ["OPENAI_API_KEY"]})

Four pillars

Package execution foundation (0.10)

Plural includes a hosted-by-default CLI, an explicit offline/private execution path, and an immutable schema-v2 execution domain. Environments own actions, typed hidden state/observation, Rewarders, resources, secrets, and runtime placement. Tasks and Verifiers are independent revisioned objects. AgentDefinition owns model, instructions, routing, and an optional Harness without binding to an Environment.

plural env init environment --name support
plural harness init harness --name support-loop
plural verifier init verifier.yaml --name correct
plural task init task.yaml --id support-1 \
  --environment environment --verifier verifier.yaml
plural benchmark init benchmark.yaml --task task.yaml
plural agent init agent.yaml --model openai/gpt-4o-mini --harness harness
plural job init job.yaml --source benchmark.yaml --agent agent.yaml
plural run job.yaml --mode eval --dry-run

Without --dry-run, an authenticated plural run validates and publishes the complete revision graph, submits a hosted Job with exact revision IDs, and follows events. Use --offline or --private to execute with a durable local Job store.

Each Task pins one Environment revision and one or more weighted Verifier revisions. A Job has a discriminated Task-or-Benchmark source and expands Agents × Tasks × attempts into Trials. Retries append TrialExecutions beneath the same Trial identity. Human verification produces awaiting_review; append-only progress events can be replayed or watched as JSON.

Execution providers are unsafe local subprocesses (explicit opt-in), hardened Docker containers, optional Daytona sandboxes, and entry-point plugins. Unsatisfiable environment requirements fail before any sandbox is created. Receipts are currently self_reported and package signatures are not verified. In train mode, Rewarders and exact TITO capture are enabled. TITO records include token IDs, output log probabilities/top log probabilities, text, assistant message, and validated input/output/observation lengths; records are stored as hashed artifacts. Eval mode disables Rewarders and TITO while still running final Verifiers. See the CLI docs, security boundaries, and known limitations.

Build a revision graph in Python

from plural import (
    AgentBinding, AgentDefinition, DeterministicVerifier, EnvironmentDefinition,
    JobSpec, TaskDefinition, TaskJobSource, WeightedVerifier,
)

environment = EnvironmentDefinition(name="support-triage")
verifier = DeterministicVerifier(name="correct", command=("python", "verify.py"))
task = TaskDefinition(
    task_id="support-1",
    instructions="Reply with the order status.",
    environment=environment,
    verifiers=(WeightedVerifier(verifier=verifier),),
    info={"order_id": "A100"},
)
job = JobSpec(
    source=TaskJobSource(task=task),
    agents=(AgentBinding(agent=AgentDefinition(name="candidate", model="openai/gpt-4o-mini")),),
)
print(job.plan().trial_count)

Docs

Follow the beginner-to-advanced documentation:

• Install and authenticate
• First offline evaluation
• Practical Python walkthrough
• Complete CLI walkthrough
• Fetch and update Plural Intel objects
• All SDK and CLI features

Detailed lifecycle, security, package schemas, and generated command/API references remain available for advanced integrations.

Development

uv sync --group dev --group docs
uv run python scripts/generate_trace_schema.py --check
uv run python scripts/generate_package_schemas.py --check
uv run python scripts/generate_cli_reference.py --check
uv run pytest -m "not live"
uv run ruff check .
uv run mypy src/plural
uv run mkdocs build --strict

License

Apache-2.0