Skip to content

DEMO_SCRIPT.md

Latest commit

 

History

History
144 lines (104 loc) · 4.64 KB

DEMO_SCRIPT.md

File metadata and controls

144 lines (104 loc) · 4.64 KB
Raw

WaveMind Demo Script

This is the public demo script for launch videos, GIFs, README snippets, and conference-style walkthroughs. Keep it reproducible and honest: no external keys, no network, no hidden setup.

30-Second Terminal Demo

Goal: show that WaveMind works immediately after cloning.

python examples/demo.py

What to say:

WaveMind stores local memory in SQLite-compatible state and returns a small recall set from a normal command-line workflow. This demo is offline and keyless.

Expected shape:

[ok] Remembered: "Andrey is a trader who tracks market breakouts."
[ok] Remembered: "Andrey prefers short practical answers about AI agents."

Query: "Andrey trader agent"
-> Result 1 (...): "..."
-> Result 2 (...): "..."

60-Second Dynamic Memory Demo

Goal: show why WaveMind is not just static nearest-neighbor search.

python examples/dynamic_memory_demo.py

What it demonstrates:

  • a corrected newer fact outranks a stale fact;
  • a temporary memory expires and is not recalled by its tag;
  • two users stay isolated by namespace;
  • the candidate index reports health against source-of-truth memory ids.

Narration:

Vector search answers "what is similar?" Agent memory also needs "what still matters?" Here the old budget is still in memory, but the corrected budget ranks first. The temporary discount code expires. Maria's namespace does not leak into Andrey's recall. The index-health line shows whether candidate search is synchronized with durable memory.

Expected shape:

WaveMind dynamic memory demo

[store]   user:andrey -> "User budget is $500."
[correct] user:andrey -> "User budget is $2000."
[expire]  user:andrey -> "Temporary discount code is ALPHA-24."
[store]   user:maria  -> "User budget is $9000."
[purge]   expired memories removed: 1

Query user:andrey: "what is the user budget?"
-> Result 1 (...): "User budget is $2000."
-> Result 2 (...): "User budget is $500."
[ok] corrected newer budget outranks the stale budget

Query user:maria: "what is the user budget?"
-> Result 1 (...): "User budget is $9000."
[ok] namespace isolation keeps Maria separate from Andrey

Query user:andrey temporary tag: "discount code"
[ok] expired temporary memory is not recalled

Index health
[ok] numpy-exact healthy=True expected=3 vectors=3

90-Second Verified Experience Demo

Goal: show the actual product breakthrough rather than only memory retrieval.

python examples/verified_experience_runtime.py

Narration:

A normal memory system can save what an agent said. WaveMind waits for an independent outcome. This cold run fails, so the failed procedure is preserved but not promoted. Repeated verified executions create a bounded, cited Experience Packet. A held-out run can reuse that packet, and an operator can inspect or roll it back. The agent learns from proof, not from its own confidence.

Show these moments:

  1. the cold attempt and independent verification;
  2. the candidate remaining in shadow before its evidence threshold;
  3. the cited Experience Packet on the next run;
  4. the successful held-out outcome;
  5. inspection and rollback.

One-Minute Product Explanation

Use this for short videos or comments:

WaveMind is a trust layer for agent memory. It captures completed tool work, waits for a test, tool, operator, or downstream state to verify the outcome, and promotes only a bounded procedure. The next run receives a compact Experience Packet with its evidence, scope, corrections, and rollback path. Vector search can still find candidates; WaveMind governs what is safe to carry forward.

What Not To Show

  • Do not claim it is faster than Chroma in static retrieval.
  • Do not claim full LoCoMo/LongMemEval answer-quality leaderboard status.
  • Do not imply the current graph is a continuous physics field.
  • Do not use private API keys or personal memory in public demos.

Recording Checklist

  1. Start from a clean terminal in the repository root.
  2. Run python examples/demo.py.
  3. Run python examples/dynamic_memory_demo.py.
  4. Run python examples/verified_experience_runtime.py.
  5. Open README and show the evidence table only after the product loop.
  6. End with the install command:
python -m pip install wavemind

Suggested Captions

  • "Vector search finds similar text. WaveMind tries to remember what still matters."
  • "Local-first memory: SQLite source of truth, vector candidates, dynamic recall."
  • "Corrections, TTL, namespaces, audit log, and index health in one dynamic memory layer."
  • "The agent learns from an independently verified outcome, not from its own confidence."
  • "Every reusable procedure keeps its source, scope, and rollback path."