Core LM Studio

Core LM Studio is a local-first Windows desktop application built on top of the Core LM full-spectrum research kit. It combines:

• an Electron + React calculator-inspired console for Core LM operations;
• a Python sidecar service that wraps the existing deterministic Core LM reference product;
• SQLite persistence for sessions, workflows, chat bus messages, ledger mirrors, replay snapshots, connector metadata, and metrics;
• an n8n-like Flow Studio for node-based ingestion, compression, Core LM execution, chat output, and outbound routing;
• deterministic mocks for external LLM/local model/REST integrations.
• sample workflows, connector configs, and programming-agent packets that can run offline through the canonical sidecar.
• provider-native Ollama usage metrics, deterministic sampling controls, structural quality evaluation, and a Compression Inspector for run metadata.
• sidecar-managed Ollama startup for real local model runs when the ollama binary is installed.
• a direct-runtime benchmark layer with strict-vs-bridge policy, direct adapter discovery, persisted benchmark profiles/runs/trials, and JSON/Markdown/CSV reports.

Core LM remains the canonical state authority. External systems are perturbation sources or outbound targets, and the global chat is an interaction bus rather than a truth store.

Studio quick start

python -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
npm install
npm run desktop:dev

Windows:

scripts\start_corelm_studio.ps1

Useful development commands:

PYTHONPATH=. python -m services.core_service.corelm_studio
PYTHONPATH=. python -m services.core_service.corelm_studio.benchmark_cli profiles
PYTHONPATH=. python -m services.core_service.corelm_studio.benchmark_cli run --profile-id builtin-runtime-conformance
npm run desktop:test
npm run desktop:build
npm run desktop:package:win
PYTHONPATH=. python -m pytest -q

Mock connectors run fully offline. For real Ollama runs, set the connector to mock=false; the sidecar probes the loopback Ollama API and starts ollama serve on demand when possible. Core LM Studio does not bundle Ollama, LM Studio, or model weights.

Primary Studio docs:

• docs/corelm_studio/quickstart.md
• docs/corelm_studio/architecture.md
• docs/corelm_studio/connectors.md
• docs/corelm_studio/workflows.md
• docs/METRICS.md
• docs/DIRECT_RUNTIME_BENCHMARK_PLAN.md
• docs/METRIC_CATALOG.md
• docs/ADAPTER_SUPPORT_MATRIX.md
• docs/STRICT_BENCHMARK_POLICY.md
• docs/BENCHMARK_PROFILES.md
• docs/DETERMINISM_INSPECTOR.md
• docs/RUNTIME_TELEMETRY.md
• docs/QUALITY_EVAL.md
• docs/COMPRESSION_INSPECTOR.md
• docs/WINDOWS_BUILD_DIRECT.md
• docs/corelm_studio/security_notes.md
• docs/corelm_studio/replay_and_ledger.md
• docs/corelm_studio/windows_build.md
• docs/implementation_plan.md
• docs/architecture_decisions.md

Core LM Full-Spectrum Kit

Core LM is a deterministic memory architecture that separates text generation from truth storage. This repository packages the project as one clean, reproducible bundle with four aligned layers:

1. Publication artifact — formal specification, reference kernel, oracle kernel, benchmark suite.
2. Archive package — source snapshot, checksums, manifest, release notes, rebuild path.
3. Sales package — one-pager, FAQ, positioning, pricing template, objection handling.
4. Local reference product — installable Python package plus CLI for structured fact ingestion, provenance, replay, and session persistence.

Claim boundary

This bundle supports these precise claims:

• publication-ready research artifact;
• archive-ready reproducible source bundle;
• sales-collateral-complete package for demos and evaluation conversations;
• complete local reference product for single-user deterministic operation.

It does not claim:

• audited security hardening;
• multi-tenant cloud operations;
• customer adoption proof;
• legal review of commercial terms.

Quick start

Local product use

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
pip install .
corelm demo --session examples/demo_session.json
corelm get --session examples/demo_session.json --branch corelm --subject project --attribute name

Development and full validation

python -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
make release-check

Common commands

# source mode
PYTHONPATH=. python -m pytest -q
PYTHONPATH=. python -m benches.runner --out reports/benchmark_latest.json --readiness-out reports/publication_readiness.json
PYTHONPATH=. python -m corelm.cli demo --session examples/demo_session.json

# editable install
pip install -e .
corelm demo --session examples/demo_session.json

# wheel build/install
python -m pip wheel . -w dist --no-deps
pip install dist/*.whl
corelm demo --session examples/demo_session.json

Repository layout

• corelm/ — kernels, baselines, product wrapper, CLI
• benches/ — frozen benchmark traces and runner
• tests/ — publication and product tests
• docs/ — publication, archive, sales, product, maintenance, and handoff docs
• reports/ — benchmark, readiness, and checksum reports
• examples/ — reproducible demo session and command walkthrough
• tools/ — cleanup and verification scripts
• apps/desktop/ — Electron, React, and TypeScript desktop shell
• services/core_service/ — Python Core LM Studio sidecar service
• packages/ — shared TypeScript contracts and UI constants
• sample_workflows/ — importable Flow Studio examples
• sample_connectors/ — mock connector configuration examples
• prompt_templates/ — programming-agent outbound packet templates
• scripts/ — Windows and cross-platform local launch helpers

Important companion docs

• docs/quickstart.md
• docs/user_guide.md
• docs/reproducibility_guide.md
• docs/limitations_and_scope.md
• docs/maintenance_guide.md
• docs/release_notes.md

Packaging status

• Publication artifact: complete
• Archive package: complete
• Sales collateral: complete
• Local reference product: complete
• Hosted production service: intentionally out of scope for this bundle