6,400+ tools. Contract-gated. Quality-evaluated.

Capability surfaces

The platform exposes capabilities through five runtime fleets, each with a distinct execution profile:

• Browser runtime — Playwright + virtual authenticators + HAR-driven journey replay. ~3,000 tools. Routed for: web automation, account provisioning, journey verification, captcha handling.
• Media worker — ffmpeg + Blender + image-generation pipelines. ~150 tools. Routed for: video assembly, terrain rendering, multi-frame visual gates.
• Data runtime — pandas, plotly, weasyprint, openpyxl, MinIO. ~250 tools. Routed for: data transformation, viz generation, PDF assembly.
• LLM worker fleet — model-router-routed across ~120k callable HF + 8 managed providers. ~3,900 tools. Routed for: text generation, code analysis, evaluation.
• API pod (in-process) — auth, storage, executor engine, renderer. Boot-critical surfaces only.

What "contract-gated" means

Every tool execution requires a PostgreSQL contract record with mandatory input fields. Readiness 100% → dispatch. No silent defaults; no missing fields. The contract is the audit trail.

What "quality-evaluated" means

External-facing tools carry mandatory quality gates: external_communications (master), format_compliance, communication_impact, research_integrity, share_page_quality, recipient_intelligence, translation_completeness. Skipping a gate requires explicit reason + approver. 494+ tools currently gated.

Capability discovery

Tool catalog is queryable: /api/v4/site-data/tools. Each tool exposes input_schema, quality_gates, category, runtime_pool, success-rate, last-run timestamp.