Examples
Choose a complete flow or a focused feature. These are the original framework examples, with their source, prerequisites and detailed implementation retained.
For hosted ATFGraph, start with your first node, adapt the example in your node project, then build, deploy and run. The Worker and Caller commands below are for a framework source checkout with the SDK, example modules and a configured graph connection; the npm installer does not include that checkout. Hosted deployment starts workers for you. Contact AI-LAB for example source access or feature enablement.
Co-consult defaults can call live suppliers and models. Follow its fixture-mode instructions before trying a deterministic demo.
Start with a complete flow
| Example | Shows | Worker | Caller |
|---|---|---|---|
| first_node/ | One traversal calling one skill, a Pydantic output schema and a synchronous handler in a thread. | python -m atfgraph.worker --module examples.python.first_node.agent:app |
python -m examples.python.first_node.call Ada |
| nested_concurrency/ | Fan-out with asyncio.gather, progress events, retries with max_attempts, cancellation propagation and capacity 1 without deadlock. | python -m atfgraph.worker --module examples.python.nested_concurrency.agent:app --capacity 1 |
python -m examples.python.nested_concurrency.call |
| sessions/ | Server-created sessions, session/actor/workflow/global memory scopes, compare-and-set updates and serialized root turns. | python -m atfgraph.worker --module examples.python.sessions.agent:app |
python -m examples.python.sessions.call |
| co_consult/ | The consultation graph: intake, three parallel desks, advisory context, evidence-checked proposals and session chat with saved-search follow-ups. Select fixture suppliers AND fixture decisions explicitly for a deterministic demo. | python -m atfgraph.worker --module examples.python.co_consult.main:app |
python -m examples.python.co_consult.client_demo |
Feature examples
| Example | Shows |
|---|---|
| features/schedule_emitter/ | Cron and interval triggers, with injected TriggerContext. |
| features/webhook_emitter/ | Authenticated webhook sources, transforms and duplicate suppression. |
| features/model_tools/ | Structured model output with capability tool calls recorded as child steps. |
| features/databricks_events/ | Emitter-driven review composing the Databricks node. |
| features/harness_delegate/ | app.harness(...) against an installed coding-agent CLI. |
| features/multimodal/ | Image, audio and file inputs; generation, transcription, usage and cost. |
| features/knowledge/ | Batch ingest and cited BM25 retrieval with extractive answers, without a model. |
| features/serverless/ | One app as a worker, a bounded invocation and a function-per-event host. |
| features/approvals/ | A durable approval pause, an out-of-band decision and checkpointed replay. |
| features/version_routing/ | Two versions of one label and the effect of pinning a caller. |
| features/access_boundaries/ | Server-enforced access policies and connector grants. |
| features/resilience/ | Retried attempts, capability budgets and cancellation with a local fixture provider. |
| features/retrieval/ | Semantic and hybrid retrieval. |
| features/realtime/ | Realtime model sessions. |
| features/offline_mesh/ | Offline mesh execution patterns. |
Emitters, harnesses and resilience
| Example | Shows |
|---|---|
| Emitters demo | Schedules and webhooks together in a complete emitter example. |
| Harness as a traversal | Coding-agent delegation within a recorded traversal. |
| Resilience verification suite | Failure handling and recovery scenarios. |
Agent systems and knowledge
| Example | Shows |
|---|---|
| Research path | A complete research workflow. |
| Docs assistant | An assistant grounded in documentation. |
| Document search | Document ingestion and search. |
| Scenario lab | Scenario-based agent workflows. |
| Retrieval quality | Inspecting retrieval quality and evidence. |
| Memory lab | State and memory behavior across calls. |
| Retrieval basics | A starting point for retrieval-backed capabilities. |
Discovery and media
| Example | Shows |
|---|---|
| Tool discovery | Discovering and calling capabilities as tools. |
| Image studio | Image workflows and outputs. |
| Speech notes | Audio and transcription workflows. |
| Browser voice | A browser client for realtime interactions. |
Integration and hosting patterns
| Example | Shows |
|---|---|
| Co-consult MCP profile | Exposing the consultation workflow through an MCP profile. |
| Two nodes on Databricks | A multi-node Databricks deployment pattern. |
| Databricks presets | Databricks-specific configuration presets. |
| Container worker & logs | A containerized worker and its logs. |
| On-demand node | Starting a node on demand. |
| Registration benchmark | A registration scale benchmark; not a hosted service capacity guarantee. |