PeaLang is an embeddable extension language for controlled rules, workflows, automation and host integration. Version 0.37 completes runtime resource accounting and structured concurrency while retaining the v0.35 standard library and v0.36 type-system foundation.
python -m pip install pealang-0.57.0-py3-none-any.whl
pea --versionpea run app.pea
pea repl
pea stdlib
pea spec
pea featuresUse std.collections, std.string, std.math, std.datetime, std.json, std.csv, std.regex, std.file, std.path, std.http, std.crypto, std.logging, std.testing, and std.concurrent. Sensitive capabilities require explicit runtime permissions.
PeaLang retains PEAC bytecode/VM, optional typing, semantic analysis, Python embedding, async VM, FastAPI integration, Java sidecar/client, TypeScript/React SDKs, DAP, package/signing/registry security and observability. See docs/USER_GUIDE.md, docs/BASICS.md, FOUNDATIONAL_ROADMAP.md, and RELEASE_NOTES.md.
Generics, structural protocol contracts, project-wide contract analysis and Result/Option/Future semantics are now part of the foundation. See docs/TYPE_SYSTEM_036.md.
Resource budgets now cover time, instructions, memory, recursion, stack, collections, output, HTTP requests and file reads. Structured concurrency adds task groups and cancellation propagation.
Use test "name" { expect condition }, pea test --filter, pea test --coverage, and pea migrate --from 0.37 --to 0.38 ..
Use pea wasm-build and pea wasm-inspect. Executable lowering remains Preview.
Run pea conformance for the 500-case interpreter/VM gate and deterministic fuzz harness. GitHub Actions defines Windows/Linux/macOS and Python 3.10-3.13 matrices.
Run pea readiness. Candidate pilots and review packs are included, but the gate intentionally remains pending until two genuine pilots are validated.
A new embeddable data foundation provides schemas, DataFrame/LazyFrame processing, a connector registry, CSV/JSON/JSONL IO and core transforms. See DATA_ROADMAP.md.
Use joins, grouped aggregations, expression-based columns and filters, null handling, pivot/unpivot, row-number windows and lazy pipelines. See docs/DATAFRAME_ENGINE_043.md.
SQLite works with the Python standard library. Adapter contracts also cover DuckDB, PostgreSQL, MySQL/MariaDB, SQL Server and Oracle through optional drivers. See docs/SQL_DATABASES_044.md.
Use the built-in REST and GraphQL connectors with runtime authentication headers, retry/backoff controls, pagination and direct DataFrame conversion. See docs/CONNECTOR_SDK_045.md.
OData and SFTP now extend the Connector SDK. See docs/CLOUD_CONNECTORS_046.md.
Amazon S3, Azure Blob Storage, ADLS Gen2 and Google Cloud Storage now share a normalized object-storage API. See docs/OBJECT_STORAGE_047.md.
Microsoft Graph/SharePoint, secret-provider abstractions, capability descriptors and certification checks extend the connector platform. See docs/ENTERPRISE_CONNECTORS_048.md.
Production runtime profiles, renewable token providers, health diagnostics, certification matrices and Graph/SharePoint pagination.
PeaLang DataFrames now provide descriptive statistics, distributions, correlation, ranking, rolling/cumulative calculations, outlier detection and dataset profiles. See docs/ANALYTICS_050.md.
Date/time resampling, shifts, period changes, expanding/EWM windows, interpolation and grouped analytics now extend PeaLang DataFrames. See docs/TIME_SERIES_051.md.
Baseline trend, moving-average and exponential-smoothing forecasts, evaluation, decomposition, grouped forecasting and chart-ready outputs extend the analytics stack. See docs/FORECASTING_052.md.
Dependency-light preprocessing, linear/logistic regression, evaluation, cross-validation indices and model persistence now extend PeaLang DataFrames. See docs/ML_FOUNDATIONS_053.md.
Fitted pipelines, cross-validation, tuning, regularization, multiclass classification, model cards and reproducibility manifests extend PeaLang ML. See docs/PRODUCTION_ML_054.md.
Model bundles, batch/audited inference, drift and fairness diagnostics, calibration, registry lifecycle and an optional FastAPI adapter extend production ML. See docs/MLOPS_055.md.
Optional Ed25519 signing/Fernet encryption, monitoring event stores, alert operations, promotion policies, canary routing and rollback controls extend the ML platform. See docs/SECURE_MLOPS_056.md.
Versioned pipeline serialization, SQLite monitoring, alert policies, deployment manifests and pea-ml operations complete the next MLOps layer. See docs/MLOPS_CLI_057.md.
Rate and concurrency controls, timeouts, structured logs, trace hooks, live reload, secure artifact loading and alert-driven rollback harden model serving. See docs/RESILIENT_SERVING_059.md.
Shared Redis-compatible state, distributed rate limiting, latency/load gates, OpenTelemetry hooks and Kubernetes resources extend resilient serving. See docs/DISTRIBUTED_SERVING_060.md.
Secure YAML manifests, non-root/read-only containers, NetworkPolicy, PDB, spread/anti-affinity, graceful draining, validation and soak tests harden distributed serving. See docs/KUBERNETES_SECURITY_061.md.
SBOM, provenance, checksums, artifact verification, non-root container packaging, CI matrix and v1 readiness gates establish the release-candidate supply chain. See docs/SUPPLY_CHAIN_062.md.
0.63 RC1 freezes feature scope and adds public API snapshots, compatibility/deprecation policy, integration gate reporting, threat modelling and upgrade guidance. See docs/API_STABILITY_POLICY.md and RC1_GATE_STATUS.md.
RC2 preserves the feature freeze and validates compatibility against RC1. See docs/RC2_STABILIZATION.md and RC2_GATE_STATUS.md.