End-User Setup Demo
Install provider context, export LangGraph runtime manifests, keep workflow state outside the LLM context, and choose tracing without adding mandatory hosted dependencies.
Provider-native root files: CLAUDE.md, AGENTS.md, and GEMINI.md.
LangGraph runtime export under .langgraph/.
Trace modes: otel, langsmith, or none.
The Problem Before
Setup is assistant-specific
Teams often copy one assistant file and lose provider-neutral behavior for the next tool.
Runtime state disappears
Long triage flows need checkpoints and durable context outside the model prompt.
Monitoring choices are unclear
Tracing should be optional, documented, and open-source-first by default.
The Solution
Install provider context at the project root, install LangGraph manifests under .langgraph/, store runtime state in a repo-local checkpoint path, and select otel, langsmith, or none for tracing.
Interactive Setup
Choose the assistant surface for the consuming repository. LangGraph remains the runtime export, not the provider.
./install.sh --provider claude --project --context-only --dry-run
./install.sh --provider codex --project --context-only --dry-run
./install.sh --provider gemini --project --context-only --dry-run
Dry-run the project root context file before writing CLAUDE.md, AGENTS.md, or GEMINI.md.
./install.sh --provider claude --project --context-only --dry-run
Install provider context at the repository root and provider assets under the provider folder.
./install.sh --provider claude --project --force
Install LangGraph graph, command, agent, and pre-dispatch context manifests.
python3 scripts/build_adapters.py
./install.sh --runtime langgraph --project --force
Keep checkpoints and durable runtime context outside the LLM context window.
NUNNERI_RUNTIME=langgraph
NUNNERI_STATE_STORE=sqlite
.nunneri/langgraph/state.sqlite
Default to OpenTelemetry for an open-source-first path. Use LangSmith only when a team wants the hosted tracing UI.
NUNNERI_TRACE_MODE=otel
NUNNERI_TRACE_MODE=langsmith
LANGSMITH_API_KEY=replace-with-your-key
NUNNERI_TRACE_MODE=none
Confirm the consumer install, LangGraph exports, and setup docs before sharing the pattern.
python3 scripts/check_consumer_install.py
python3 scripts/check_langgraph_exports.py
python3 scripts/check_user_setup_docs.py