NovaFabric Tutorials

Plain-English guides for engineers at every level. Read in any order.

Start here

Tutorial What you learn
Why NovaFabric? The problem it solves, with five concrete scenarios using real testbench workloads
Prove a run to an auditor The scenario the project exists for: capture and seal a run, archive it, then six months later verify it is unmodified, replay it offline, and hand over an Evidence Bundle — with an explicit section on what it does not prove
Getting Started The 15-minute first run: install, capture, validate, replay, diff, lineage, seal
Feature tour Hands-on tour of every capability — proxies, all providers, LangChain, evidence bundles, knowledge graph, capture-level policy, GDPR erasure, dashboard reports, topology, compliance exports, the zero-token eval loop, intervention replay, the Accountability Spine (energy / ledger / safety case), incident tracking, plus the Langfuse-parity cohort: prompt lifecycle and labels, offline analytics (query / views / trends / pricing), sessions and the execution graph, and the team evaluation workflow
5-minute black-box recorder demo End-to-end demo: capture → validate → scan-secrets → replay → diff → lineage → verify; no live API key required

Go deeper

Tutorial What you learn
How capture works What HTTP is, what requests and aiohttp are, how monkey-patching intercepts LLM calls without code changes, the URL registry, all four capture modes
Multi-agent capture Same-process vs separate-process topologies, how to capture each, MCP sub-agents, the parent/child linking design (ADR-0032)
Cluster scale — 1,000,000 agents Where capsules live (always filesystem), why the current single-cluster design doesn't reach 1M agents, the self-contained rule, the four-layer architecture, the three hard problems (storage, lineage, identity), and the phased build plan

Compare and evaluate

Tutorial What you learn
NovaFabric vs Langfuse Where they overlap, where they differ, honest assessment of what each does better