Same output. Less compute.
A side-by-side demonstration comparing standard LLM token usage against the AIEP incremental reasoning substrate — with live efficiency metrics.
| Column | What it represents |
|---|---|
| Standard | Stateless re-derivation — how many tokens a standard LLM call would consume starting from scratch |
| AIEP | Incremental recall substrate — actual tokens used after pulling from committed prior reasoning |
The efficiency panel (expandable) shows the full P117 parametric unburdening breakdown: reused steps, fresh steps, tokens avoided, HAIR chain recalls, source diversity score, and the raw evidence artefacts the substrate drew from. When the HAIR cross-session recall overlay fires, the efficiency summary line appends the count of recalled artefacts.
The substrate also exposes a hash verification endpoint — any response hash can be checked against the committed ledger without authentication.
- Open
index.htmlin any browser — no build step required. - Set your PIEA API endpoint on
<body data-api-base="https://your-piea.workers.dev">.
Without this, it defaults tohttp://localhost:8788(local development). - Enter a question and click Run.
If you have a PIEA worker running locally on port 8788, the demo works with no configuration. The demo calls:
POST /api/ask—{ question, session_id }→{ answer, efficiency, evidence_rail, usage, llm_usage, source_diversity_score, connection_rail }GET /api/stats—{ tokens_saved_total, … }POST /api/recall/verify—{ hash: "<64-char SHA-256>" }→{ verified, ledger_record }(unauthenticated)
{
"reuse_pct": 68,
"reused_steps": 4,
"fresh_steps": 2,
"tokens_avoided": 3200,
"baseline_tokens": 4700,
"reduction_ratio": 0.68,
"hair_recalls": 4
}hair_recalls is the count of prior artefacts injected by the HAIR cross-session recall overlay. When hair_recalls > 0, tokens_avoided = hair_recalls × 800; otherwise tokens_avoided = reused_steps × 500.
index.html — self-contained demo, no dependencies
README.md — this file
LICENSE — Apache 2.0
- aiep.dev — project home
- AIEP Live — real-time operating metrics dashboard
- AIEP Utilities — shared tooling for the AIEP protocol
Apache 2.0 — see LICENSE.