Long-Doc Chat - AG2

07 · Long-Doc Chat

A "remember the last words" chat exercise that stress-tests the assembly chain. Three policies compose to control exactly what the LLM sees on each call — drop non-conversation events, hard-cap to the last 6 events, then enforce a token budget. A separate TailWindowCompact strategy keeps the underlying stream history small too. Watch the compaction events fire as the conversation grows.

What it covers

Composing multiple AssemblyPolicy instances in the order they should apply.
ConversationPolicy to drop lifecycle/internal events from the LLM's view.
SlidingWindowPolicy(max_events=N, transparent=True) for a hard event-count cap.
TokenBudgetPolicy(max_tokens=N) as a belt-and-braces secondary cap.
Combining the assembly chain with a KnowledgeConfig that wires TailWindowCompact for stream-history compaction (different from assembly!).
Watching CompactionCompleted events fire from outside the Agent.

Primitives covered

assembly=[ConversationPolicy(), SlidingWindowPolicy(...), TokenBudgetPolicy(...)]
KnowledgeConfig with compact=TailWindowCompact(...) + compact_trigger=CompactTrigger(...)
MemoryKnowledgeStore (in-memory variant of the knowledge store)
CompactionCompleted lifecycle event

Source

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 """07 · Long-doc chat — composing assembly policies Shows the assembly chain in action. Three policies compose in order: 1. ``ConversationPolicy`` — drops every event that isn't conversation or tool traffic (no lifecycle noise reaches the LLM). 2. ``SlidingWindowPolicy(max_events=6)`` — hard-caps the number of events forwarded to the LLM, so history can't grow unbounded. 3. ``TokenBudgetPolicy(max_tokens=2000)`` — character-based secondary cap, belt-and-braces against one huge event blowing the budget. Also pairs the assembly chain with ``TailWindowCompact`` so the agent's stream history itself (not just the view into it) is kept small. Run:: .venv/bin/python 07_long_doc_chat.py """ import asyncio from ag2 import Agent, KnowledgeConfig from ag2.compact import CompactTrigger, TailWindowCompact from ag2.config import GeminiConfig from ag2.events import CompactionCompleted from ag2.knowledge import MemoryKnowledgeStore from ag2.policies import ( ConversationPolicy, SlidingWindowPolicy, TokenBudgetPolicy, ) from ag2.stream import MemoryStream def section(title: str) -> None: print(f"\n── {title} ───") QUESTIONS = [ "Remember the word 'oak'.", "Remember the word 'river'.", "Remember the word 'lantern'.", "Remember the word 'sable'.", "Remember the word 'quartz'.", "Name the three most recent words I asked you to remember.", ] async def main() -> None: config = GeminiConfig(model="gemini-3-flash-preview", temperature=0) store = MemoryKnowledgeStore() compactions: list[CompactionCompleted] = [] stream = MemoryStream() stream.where(CompactionCompleted).subscribe(lambda e: compactions.append(e)) agent = Agent( "lexicon", prompt=( "Be very terse — one short sentence per reply. " "Answer directly without calling any tools." ), config=config, assembly=[ ConversationPolicy(), SlidingWindowPolicy(max_events=6, transparent=True), TokenBudgetPolicy(max_tokens=2000), ], knowledge=KnowledgeConfig( store=store, compact=TailWindowCompact(target=4), compact_trigger=CompactTrigger(max_events=8), ), ) section("Long-doc chat — assembly policies trim what the LLM actually sees") reply = await agent.ask(QUESTIONS[0], stream=stream) print(f"Q1> {QUESTIONS[0]}") print(f"A1> {reply.body}") for i, q in enumerate(QUESTIONS[1:], start=2): reply = await reply.ask(q) print(f"Q{i}> {q}") print(f"A{i}> {reply.body}") print() print(f"Compactions fired during run: {len(compactions)}") for c in compactions: print(f" - {c.strategy}: {c.events_before} → {c.events_after} events") if __name__ == "__main__": asyncio.run(main())


<br> 1<br> 2<br> 3<br> 4<br> 5<br> 6<br> 7<br> 8<br> 9<br>10<br>11<br>12<br>13<br>14<br>15<br>16<br>17<br>18<br>19<br>20<br>21<br>22<br>23<br>24<br>25<br>26<br>27<br>28<br>29<br>30<br>31<br>32<br>33<br>34<br>35<br>36<br>37<br>38<br>39<br>40<br>41<br>42<br>43<br>44<br>45<br>46<br>47<br>48<br>49<br>50<br>51<br>52<br>53<br>54<br>55<br>56<br>57<br>58<br>59<br>60<br>61<br>62<br>63<br>64<br>65<br>66<br>67<br>68<br>69<br>70<br>71<br>72<br>73<br>74<br>75<br>76<br>77<br>78<br>79<br>80<br>81<br>82<br>83<br>84<br>85<br>86<br>87<br>88<br>89<br>90<br>	<br>"""07 · Long-doc chat — composing assembly policies<br>Shows the assembly chain in action. Three policies compose in order:<br>1. ``ConversationPolicy`` — drops every event that isn't conversation or<br> tool traffic (no lifecycle noise reaches the LLM).<br>2. ``SlidingWindowPolicy(max_events=6)`` — hard-caps the number of events<br> forwarded to the LLM, so history can't grow unbounded.<br>3. ``TokenBudgetPolicy(max_tokens=2000)`` — character-based secondary cap,<br> belt-and-braces against one huge event blowing the budget.<br>Also pairs the assembly chain with ``TailWindowCompact`` so the agent's<br>stream history itself (not just the view into it) is kept small.<br>Run::<br> .venv/bin/python 07_long_doc_chat.py<br>"""<br>import asyncio<br>from ag2 import Agent, KnowledgeConfig<br>from ag2.compact import CompactTrigger, TailWindowCompact<br>from ag2.config import GeminiConfig<br>from ag2.events import CompactionCompleted<br>from ag2.knowledge import MemoryKnowledgeStore<br>from ag2.policies import (<br> ConversationPolicy,<br> SlidingWindowPolicy,<br> TokenBudgetPolicy,<br>)<br>from ag2.stream import MemoryStream<br>def section(title: str) -> None:<br> print(f"\n── {title} ───")<br>QUESTIONS = [<br> "Remember the word 'oak'.",<br> "Remember the word 'river'.",<br> "Remember the word 'lantern'.",<br> "Remember the word 'sable'.",<br> "Remember the word 'quartz'.",<br> "Name the three most recent words I asked you to remember.",<br>]<br>async def main() -> None:<br> config = GeminiConfig(model="gemini-3-flash-preview", temperature=0)<br> store = MemoryKnowledgeStore()<br> compactions: list[CompactionCompleted] = []<br> stream = MemoryStream()<br> stream.where(CompactionCompleted).subscribe(lambda e: compactions.append(e))<br> agent = Agent(<br> "lexicon",<br> prompt=(<br> "Be very terse — one short sentence per reply. "<br> "Answer directly without calling any tools."<br> ),<br> config=config,<br> assembly=[<br> ConversationPolicy(),<br> SlidingWindowPolicy(max_events=6, transparent=True),<br> TokenBudgetPolicy(max_tokens=2000),<br> ],<br> knowledge=KnowledgeConfig(<br> store=store,<br> compact=TailWindowCompact(target=4),<br> compact_trigger=CompactTrigger(max_events=8),<br> ),<br> )<br> section("Long-doc chat — assembly policies trim what the LLM actually sees")<br> reply = await agent.ask(QUESTIONS[0], stream=stream)<br> print(f"Q1> {QUESTIONS[0]}")<br> print(f"A1> {reply.body}")<br> for i, q in enumerate(QUESTIONS[1:], start=2):<br> reply = await reply.ask(q)<br> print(f"Q{i}> {q}")<br> print(f"A{i}> {reply.body}")<br> print()<br> print(f"Compactions fired during run: {len(compactions)}")<br> for c in compactions:<br> print(f" - {c.strategy}: {c.events_before} → {c.events_after} events")<br>if __name__ == "__main__":<br> asyncio.run(main())<br>