Use case · Voice AI

Voice agents get one round trip, and it has to be the right one

In a chat, a slow answer is a spinner. On a call, it is dead air, and the caller starts talking over the agent. Voice leaves no budget for retrieving the wrong thing.

The mixture

01 · Persistent memory: Uses
02 · Current context: Uses
03 · Traceable decisions: Uses
04 · Shared context: Light
05 · Scoped retrieval: Leads

01 · What breaks

Latency is the whole product

Every retrieval choice that is invisible in text is audible on a call.

01
Every extra lookup is audible
Agents that search, reflect and search again feel fine in text. On a call, the second lookup is the moment the caller says "hello?"
02
Wide retrieval is slow retrieval
Searching the whole knowledge base for every utterance spends latency the conversation does not have, and most of what comes back is irrelevant to this caller.
03
Callers repeat themselves
The caller explained the problem yesterday. Today's agent asks again, and a caller who has to repeat themselves asks for a human.

02 · What it does

Why voice AI teams choose Alchemyst

Verified capabilities built for enterprises deploying voice AI at scale.

Unmatched context speed

Complex conversation context is processed in 170ms, so voice agents respond instantly, without latency delays.

Superior context awareness

The highest memory F1 score in the industry (0.76) keeps conversation continuity precise.

Enterprise economics

Save 83% on costs while delivering 12x more performance value per dollar than competing solutions.

Verified and transparent

Tested in December 2025 on publicly available benchmarks. No hidden claims, only results.

Pareto frontier performance

Positioned at the efficiency frontier: the best balance of cost and performance available today.

Easy integration

REST APIs and SDKs that integrate with your existing voice stack in hours, not months.

03 · Performance

170ms latency. Best-in-class efficiency.

Benchmark-proven performance for voice, tested in December 2025 on publicly available benchmarks. This is how the Alchemyst context engine defines the new Pareto frontier for voice AI.

170ms
P50 latency: 12x
Value ratio: 83%
Cost savings: 0.76
Memory F1 score

Metric	Competitors	Alchemyst
P50 latency	500-800ms	170ms
Memory F1 score	0.45-0.58	0.76
Value ratio (performance / cost)	1x-2x	12x
Cost savings	Baseline	83% reduction
Setup time	2-4 weeks	Under 24 hours

04 · What it's made of

Mostly scoped retrieval. Then three others.

Every application on Alchemyst is a different mixture of the same five jobs. That mixture is what makes this a different piece of software from the one next to it, even though the API underneath is identical.

Leads · Scoped retrieval

The nearest match stops winning

Scopes are set before the call connects: this caller, this account, this intent. Each turn runs one narrow search in fast mode, so what comes back is small, relevant and inside the latency budget.

One narrow search per turn.
Scopes set before the first word.
Fast mode for in-call lookups.

Uses · Persistent memory

Yesterday's call, remembered

What the caller said last time is written back and read before they speak.

Uses · Current context

Today's hours and offers

Superseded scripts and promotions are subtracted, so the agent never offers what has expired.

Uses · Traceable decisions

Why the agent said that

Every spoken answer carries the sources behind it, for QA review.

That is four of the five. The fifth, shared context (what one agent learns, the next one already knows, on the same definitions), is what leads in Sales agents, Customer success and Enterprise operations instead. Same API, different mixture.

05 · Where it applies

Transform any voice use case

Deliver the same context engine across your entire voice AI stack.

Sales & outbound calling
Qualify leads and close deals with context-aware agents that understand customer history and objections.
3x faster call completion
Customer support
Resolve issues faster with instant context retrieval on customer history and preferences.
40% faster resolution
Collections & reminders
Intelligent due reminders that understand payment history and customer circumstances.
25% improvement in recovery rates
Inbound call handling
Route and handle incoming calls intelligently, with complete context on the first ring.
Instant smart routing

06 · What feeds it

The sources you already have

Works with your voice stack: Plivo, Twilio and any existing telephony, with OpenAI or Anthropic models. The LiveKit plugin adds persistent cross-session memory, and production-ready APIs deploy in under 24 hours.

LiveKitPlivoTwilioOpenAIAnthropicCall transcriptsKnowledge base articlesCRMOrder status (PostgreSQL)

01 · Ingest

Scope what you ingest

Every document lands with a groupName: the sets it belongs to. Those sets are what retrieval intersects later, so the structure you choose here is the precision you get there.

context.add

import AlchemystAI from "@alchemystai/sdk";const client = new AlchemystAI(); // reads ALCHEMYST_AI_API_KEYawait client.v1.context.add({  context_type: "resource",  scope: "internal",  source: "kb",  documents: [{    content: "Standard delivery: 2 to 4 business days. Same-day delivery in Bengaluru and Mumbai for orders placed before 1pm.",  }],  metadata: {    fileName: "delivery-windows.md",    groupName: ["voice", "support", "delivery"],   // the sets this belongs to  },});

02 · Write

Write what the caller needs next time

The call ends and the transcript is long. Keep the part the next call needs: the issue, what was promised, what is still open.

context.memory.add

await client.v1.context.memory.add({  sessionId: "caller_4410",  contents: [{    role: "assistant",    content: "Order #88213 reported missing. Promised a callback by 6pm with a courier update.",  }],  metadata: { groupName: ["voice", "support", "caller_4410"] },});

03 · Search

Search before speaking

Search intersects the scopes, subtracts superseded and duplicate content, and ranks what survives. Only that reaches the model, and the whole decision is recorded as a Context Trace.

context.search

const { contexts } = await client.v1.context.search({  query: "Where is my order?",  scope: "internal",  mode: "fast",  similarity_threshold: 0.8,  minimum_similarity_threshold: 0.5,  metadata: { groupName: ["voice", "support", "delivery"] },   // ∩ narrow scope});// − superseded, deduplicated → ranked → into the window// Every search is recorded as a Context Trace.const reply = await llm.respond(message, { context: contexts });

npm install @alchemystai/sdk or pip install alchemystai, both ship the same client. Full reference in the docs.

07 · Where it runs

Every network hop is audible

Voice agents feel the distance between the context layer and the model. Keep retrieval close to inference, keep scopes tight, and keep caller memory deletable when a customer asks.

Security & compliance

Deployment

Managed cloud
Encrypted in transit and at rest, isolated per organization, and scoped at write time.
Dedicated infrastructure
Enterprise
Single-tenant, with VPC peering when your data cannot share a network boundary.
Self-hosted
On-premise
Run the context layer on your own infrastructure, with OpenTelemetry for observability.

08 · Questions

Frequently asked questions

What makes Alchemyst different?

Alchemyst is the only AI context engine with verified, publicly available benchmarks. Our 170ms P50 latency and 0.76 memory F1 score are tested and proven, not marketing claims.

How much can we save?

Alchemyst delivers 12x more value per dollar with 83% cost savings versus traditional engines. A typical enterprise saves $50K to $500K a year, depending on call volume.

How quickly can we deploy?

In less than 24 hours. The APIs are designed for rapid integration with Twilio, Plivo or any custom telephony system.

Is there a setup cost or long-term contract?

No setup fees and no contracts. You only pay for what you use. Start free with a 30-day trial and full production access.

09 · The rest