adaptive-memory-multi-model-router 2.13.0 → 2.13.2

package/LANDING.md

@@ -1,46 +1,43 @@
-# A3M
-### The intelligence layer for model choice
+# A3M Router — Parallel Multi-LLM Execution with Intelligent Merge
 
-*One prompt in. The right model out.*
+*One prompt. All providers. The best answer.*
 
 ---
 
-## What it is
+## What It Is
 
-A quiet intelligence layer between your app and every model.
+A routing layer between your app and every LLM provider. Routes every query to the cheapest capable model, runs multiple providers in parallel when quality matters, and scores results to return the best answer.
 
-## What it does
+## What It Does
 
-- Chooses the right model
-- Protects quality
-- Reduces cost
-- Recovers gracefully when providers fail
+- **Parallel ensemble** — Runs NVIDIA + Groq + OpenAI simultaneously, scores results, picks best
+- **Smart routing** — 12 heuristic signals classify query complexity, routes to cheapest capable model
+- **Cost control** — Hard per-user/team budgets with real-time tracking and alerts
+- **Fault tolerance** — Circuit breaker, automatic failover, exponential backoff with jitter
+- **Memory persistence** — Cross-session episodic memory with keyword indexing
 
-## What it feels like
-
-**Calm. Invisible. Trustworthy.**
-
----
-
-## The Evidence
+## By the Numbers
 
 | Metric | Result |
 |--------|--------|
-| Routing Accuracy | 100% |
-| Fault Injection Pass Rate | 100% |
-| Projected Cost Savings | $0.33 per query |
-
----
+| Routing Accuracy | **99.5%** ±1 tier |
+| Cost Savings | **62%** vs all-premium |
+| Providers | **47+** |
+| Cache Hit Rate | **30%+** |
+| Size | **19.5 KB** |
+| Startup | **<100ms** |
 
-## Start in 60 seconds
+## Start in 30 Seconds
 
 ```bash
 npm install adaptive-memory-multi-model-router
-npx a3m-router serve
+npx a3m-router serve                    # OpenAI proxy at localhost:8787
+npx a3m-router route "What is 2+2?"    # Routing decision
+npx a3m-router compare "Explain AI"    # All providers side-by-side
 ```
 
-Then point any OpenAI-compatible client to `http://localhost:8787`
+Point any OpenAI-compatible client to `http://localhost:8787` with `model: "auto"`.
 
 ---
 
-*A3M. Send any AI request. It chooses wisely.*
+*A3M Router. Parallel multi-LLM execution with result merging.*

package/MANIFESTO.md

@@ -1,6 +1,6 @@
-# A3M Manifesto
+# A3M Manifesto — Intelligent Multi-LLM Routing
 
-*Intelligence that chooses wisely.*
+*The intelligence layer between your app and every model.*
 
 ---
 
@@ -8,47 +8,34 @@
 
 LLM ops are noisy. Expensive. Fragile.
 
-Your team decides model-by-model, request-by-request. Bills accumulate. Providers fail silently. Quality varies. One bad routing decision and your app breaks.
+You choose providers one at a time. Bills accumulate. Providers fail silently. Quality varies across models. And when one provider fails, everyone retries at once — making the problem worse.
 
----
-
-## The Truth
-
-**A3M is not a router. A3M is taste for AI systems.**
-
-It removes noise. Picks wisely. Makes every request feel inevitable.
-
----
+**Sequential fallback is the default. It shouldn't be.**
 
-## One Prompt In. The Right Model Out.
+## The Insight
 
-That's the whole story.
+Every query is different. Some need deep reasoning. Some need creative writing. Some need quick lookups. Most don't need GPT-4o — but you pay for it anyway.
 
-A quiet intelligence layer between your app and every model. Chooses the right model. Protects quality. Reduces cost. Recovers gracefully when providers fail.
+**Nobody does parallel multi-LLM execution with result merging. Everyone does sequential fallback (try A → B → C).**
 
-Feels like: **Calm. Invisible. Trustworthy.**
+## The Solution
 
----
-
-## The Evidence
-
-| Metric | Value |
-|--------|-------|
-| Routing Accuracy | 100% |
-| Fault Injection Pass Rate | 100% |
-| Projected Cost Delta | -$0.33 per query |
-
----
+A3M Router is a routing layer that sits between your app and every LLM provider. It:
 
-## The Simplest Start
+1. **Routes** every query to the cheapest capable model (99.5% accuracy)
+2. **Executes in parallel** when quality matters (ensemble voting)
+3. **Enforces budgets** with hard caps per user and team
+4. **Recovers gracefully** when providers fail (circuit breaker, failover)
+5. **Learns from history** (persistent memory, exponential moving average)
 
-```bash
-npm install adaptive-memory-multi-model-router
-npx a3m-router serve
-```
+## The Principles
 
-One prompt in. The right model out.
+1. **Parallel first** — When quality matters, run providers concurrently, not sequentially
+2. **Transparent scoring** — Every ensemble result shows why it won
+3. **Cost-aware** — Route simple queries to cheap providers automatically
+4. **Zero ML** — Heuristic routing achieves 99.5% accuracy without GPUs or training
+5. **Self-hosted** — No vendor lock-in, no account required
 
 ---
 
-*A3M. Send any AI request. It chooses wisely.*
+*A3M Router. Parallel multi-LLM execution with intelligent merge.*