Capabilities

Control-rich defaults for teams shipping real SLM systems.

BrewSLM gives immediate productivity through autopilot presets while preserving full control over data adapters, runtime preflight, training behavior, and deployment packaging.

Capability Clusters

What you can do end-to-end

Data Intake + Mapping

Ingest local, URL, HF, Kaggle sources in one interface.
Generate canonical schema mappings and reusable profiles.
Apply validation checks before training visibility.

Contract + Runtime Preflight

Validate task and model compatibility constraints.
Check runtime dependencies and memory fit before launch.
Choose strict/fail-open policy per environment.

Benchmark Intelligence

Run sampled candidate comparisons from Training UI and model-selection APIs.
Rank by quality, latency, and budget dimensions.
Persist benchmark traces for reproducible selection.

Training Orchestration

Launch one-click autopilot runs from CLI or Wizard UI.
Track checkpoints, logs, retries, and cancellation state.
Tune plan choices with intent, runtime, and profile overrides.

Artifact Packaging

Bundle weights, config, metrics, and run metadata.
Export to serving targets such as vLLM-compatible flows.
Version and verify bundles using checksums.

Operational Governance

Enforce evaluation gates and registry promotion checks.
Track run provenance across data and model versions.
Standardize handoff between AI and platform teams.

Control Depth

How much control each mode gives you

Control Area	Autopilot Default	Manual Override	Outcome
Schema Mapping	Suggested canonical mapping	Edit mapping and validation rules	Faster setup with traceable field decisions
Model Selection	Benchmark-informed shortlist	Pin exact model and constraints	Balanced quality, cost, and consistency
Training Config	Generated train plan	Tune optimizer, schedule, stop policy	Safe baseline with targeted optimization
Runtime Placement	Recommended local/cloud placement	Force local or burst by policy	Operational flexibility by workload
Release Gating	Default quality checks	Define custom promotion thresholds	Reduced regression risk on deployment

Automation Playbooks

Starter patterns teams automate in the first month

Nightly Runtime Readiness

Check GPU/dependency/secret readiness before scheduled automation starts.

$ ./brewslm doctor --project 1

Weekly Autopilot Retrain

Kick off a guided one-click run from intent text with project defaults.

$ ./brewslm train --project 1 --autopilot --one-click --intent "Refresh support model"

Release Candidate Export

Tune for device constraints, then package an export candidate for serving.

$ ./brewslm optimize --project 1 --target mobile_iphone15

$ ./brewslm export --project 1 --format huggingface --target vllm

Capability Test Drive

Validate the core stack with one short run

$ ./brewslm project create --name demo-slm --template general

$ ./brewslm dataset import --project 1 --sample general-chat-v1

$ ./brewslm preflight --project 1 --task causal_lm

$ ./brewslm train --project 1 --autopilot --one-click

$ ./brewslm export --project 1 --format huggingface --target vllm