Advanced Fast Line Pro: streaming and completion
Optional track — for those who already completed basic Fastline (
action_fastline_pro) and want to understand production-ready patterns. Not required to complete the core track.
Required permissions and access
| Module / setting | Why you need it |
|---|---|
| Scenario Builder | Reverse engineering existing sections |
| Fast Line Pro | Agent vs Chatbot, streaming |
| Runs / Chat preview | Trace edges and variables |
| Access to a production/dev scenario (optional) | RE lab on a real flow |
Business context
In production scenarios, Fastline often uses not only action_fastline_pro, but also completion nodes, streaming replies, and multi-step agent chains. This Use Case teaches you to read and understand such patterns through reverse engineering — without a mandatory from-scratch build.
Expected result
- You understand the difference between Chatbot and Agent in Fast Line Pro
- You can explain where
fastline_completion/ streaming appear in a scenario - Reverse engineering: 10 control questions for a production section (or RE lab)
- Self-check without escalating to a mentor
Key concepts
| Pattern | Purpose | Core track |
|---|---|---|
action_fastline_pro |
FAQ, categorizer | Use Case 1, 10 |
fastline_completion |
Completing a multi-turn agent loop | Optional |
| Streaming | Partial reply in the channel before full completion | Optional |
| Agent (Fast Line Pro) | Tools + advanced modes | When tools are needed, not “just FAQ” |
| Symptom | Solution |
|---|---|
| Unclear, Chatbot or Agent | FAQ / categorizer → Chatbot; tools / streaming → Agent (step 5 below) |
| Confusion about streaming / completion | This optional Use Case |
Step-by-step implementation (reverse engineering)
Step 1. Baseline — your Use Case 1
Open the Use Case 1 flow in Scenario Builder:
| Question | Your answer (write it down) |
|---|---|
Which agent_name? |
|
| Which variable for the reply? | |
How many edges on action_fastline_pro? |
|
| What happens without a fallback edge? |
Verify the answers in Runs.
Step 2. Find completion in the documentation
- Open the actions reference.
- Find
fastline_completion(or the current equivalent on your instance). - Write down: input parameters, output edges, difference from
action_fastline_pro.
| Criterion | action_fastline_pro |
fastline_completion |
|---|---|---|
| Typical use case | One request → reply | Completing an agent session |
| Streaming | Usually no | May be related |
| Core training | Yes | Optional |
Step 3. RE lab — production section
Use the reverse-engineering-lab or a production section you have read-only access to:
- Find all Fastline nodes (pro, completion, others).
- Draw a text data flow (Start → … → End).
- For each Action node — list of edges (success, error, fallback).
10 control questions (example):
- Where is
conversation_idstored? - Which edge on an empty AI reply?
- Is there a streaming parameter in node_params?
- How many Fastline nodes are in the chain?
- Where is handoff to an operator?
- Which variables land in constants after completion?
- Is there a retry loop?
- Fallback at section or node level?
- Agent or Chatbot in Fast Line Pro?
- What happens on timeout?
Step 4. Streaming (theory + Runs)
Streaming in a channel means: the customer sees parts of the reply before generation finishes.
| Self-check | Action |
|---|---|
| Find streaming in docs | Fast Line Pro |
| Compare with non-streaming Runs | One Run with pro without streaming |
| Record the UX difference | In the result template — 2–3 sentences |
If streaming is unavailable on the training instance — record that in the self-check (not a blocker).
Step 5. Chatbot vs Agent (capability-based decision)
| Type | What it can do | When to choose |
|---|---|---|
| Chatbot | Replies from KB/prompt; no tools | FAQ, categorizer, typical action_fastline_pro |
| Agent | Same + Tools and advanced modes (streaming, etc.) | When AI must call tools / external actions |
Rule: do not pick Agent “just in case”. If you only need KB replies — Chatbot is enough (as in Use Case 1).
Troubleshooting
| Symptom | What to check |
|---|---|
| Completion is not called | Node order: pro → completion; conversation_id |
| Streaming is “choppy” | Channel (TG vs widget); streaming docs |
| RE lab is unclear | Start from Use Case 1 baseline; smaller section |
Self-check
Independent task
In a production scenario (or RE lab), find one node that can be replaced with action_fastline_pro + fallback without losing the happy path — justify in 5 sentences.