REUSABLE APPLIED INTELLIGENCE INFRASTRUCTURE
Not just model calls. Evaluate. Replay. Simulate. Deploy. Observe. Govern.
Built inside the 6away ecosystem. Designed to be reused outside it.
Currently powering GCP Pro, Coherence Guru, Rotalink, 6degrees, Ritual Guru and the wider 6away ecosystem.
THE SHAPE OF AI TODAY
Prompt → Model → Hope
A full lifecycle — measurable at every step.
CAPABILITY PLATFORM
Apps discover, require, and call Engine capabilities without rebuilding the AI plumbing underneath each one. The same routing, budget, retry, and observability machinery covers every capability — including ones added next quarter.
Transcribe audio with Whisper. Optionally extract structured fields from the transcript.
Model routing · multipart upload · retries · cost ceiling · usage logging
Audio capture UX · transcript storage · pedagogy / output use
Turn a natural-language search into a structured intent + concept bag, then rank caller-supplied candidate profiles.
Intent expansion · ranking · schema validation · output sanitisation
CRM data · candidate text assembly · UI for results
Classify GCP Net Variance into a continuous coherence state (CS / DD / IS / AT / SS / CL / SH / FA / DS) with confidence + direction + phase.
AI classification · longitudinal memory · schema-strict output · fallback chain
Net-variance computation · chart rendering · alerts · the trading decision
Structured company intelligence extracted from web content and open-source signals.
Scraping orchestration · AI extraction · structured output shape
CRM persistence · contact-company linking · ranking inside your product
Compress free-form text into a structured summary plus key-points array. Eval-handler backed for side-by-side model comparison.
Model selection · eval-supported quality bench-off · retries
Source text · presentation · where the summary lives in the product
Extract structured contact information from email, LinkedIn, web, documents, Telegram or WhatsApp content the caller supplies.
Source-specific prompts · schema validation · output sanitisation · per-source routing
Source data · contact storage · dedup · ownership rules
The reverse channel. Apps periodically send 'this is what happened' under a registered domain adapter — recommendation accepted, follow-up completed, ritual finished, habit logged. The Engine validates and persists; outcome and correction land alongside to become evidence.
Adapter registry · event validation · workspace-scoped persistence · admin observability
Which signals to send · payload shape inside the adapter contract · when to fire
Capabilities run through a governed chain as they mature onto it: provider routing, evaluation against alternative models, replay against historical traffic, and shadow / canary deployment.
Provider routing · fallback · workspace budget · per-call observability · canary auto-stop
Nothing — these are infrastructure-level guarantees every consumer inherits for free
ECOSYSTEM FEEDBACK LOOP
The loop is the point. Real workload validates capabilities; observations turn into evidence; evidence sharpens the surfaces every app consumes. Deterministic today; the learning layer is under construction.
Every application does three jobs.
Apps own their customer, their product, their pricing. The Engine never gets between an app and its users.
Real workload validates Engine capabilities. Production traffic exposes what evals can't see.
Observations from real usage become the evidence the Engine learns to understand — deterministically first, intelligently later.
DOMAIN ADAPTERS
Each product connects through a domain adapter — a small declarative bundle that tells the Engine what signals the product emits and what privacy posture applies. The Engine doesn't care that one app is relationship intelligence and another is daily rituals; it sees the same observation shape across all of them.
WHAT AN ADAPTER DECLARES
Relationship and meeting outcomes. The first adapter wired in production.
Market coherence-state predictions and the actual subsequent behaviour.
Lesson outcomes — answer submissions, mastery transitions, session lifecycle.
Workout outcomes and load progressions — sanity check for non-conversational domains.
The densest behavioural-state stream: rituals, meditation, reflection, mood / energy / focus / sleep / HRV, habits.
OBSERVATION → EVIDENCE
An observation is the raw event: a recommendation accepted, a habit completed, a ritual finished. Evidence is what that event becomes once the outcome arrives — and once any correction the user made gets attached to it. The Engine accumulates evidence; aggregation comes first, intelligence comes later.
Deterministic today. The learning layer is under construction; AI enters in a future, named phase.
Today the Engine accumulates evidence. Aggregation and pattern discovery are deterministic and explicit. The learning layer is under construction; AI enters in a future, named phase.
Every observation is workspace-scoped. Adapter declarations pin sharedAcrossWorkspaces = false. Cross-domain inference, if it ships, gets its own consent surface.
Each product owns its CRM data, its UX, its decisions. The Engine never reads an app's database, never renders product UI, never persists caller-owned business data.
GOVERNANCE · SAFETY · SUBSTRATE
This is the substrate capabilities run on. Provider routing, evaluation, replay, simulation, shadow-then-canary deployment with auto-stop. Memory, a status contract, a typed SDK.
Route across Claude, GPT, Gemini, Infercom, MiniMax and local models without changing application code.
Compare models on structured tasks. Measure quality, latency, reliability, schema validity and cost — before production.
Replay real production requests against new models. See exactly what changes before deployment.
Test routing strategies against historical traffic. Estimate cost, latency, reliability and fallback behaviour without touching production.
Run a candidate model alongside the incumbent. Promote only when the candidate proves itself on real workload.
Canary deployments freeze themselves when divergence, error rate, or cost cross declared thresholds. No operator action required.
Track behaviour over time. Understand state persistence, transitions, volatility and provider disagreement — without giving memory control of routing.
App-facing health + capability contract. Workspace-aware. Surfaces degraded flags so callers can render fallbacks before failure.
Hand-written TypeScript client — no zod / drizzle dependencies. Typed helpers for stable contracts. Copy-pasteable into any app repo.
FOR BUILDERS
The Engine is provisioned per workspace. Same API key serves every capability. New capabilities show up in your catalog the moment they ship.
PHILOSOPHY
Most AI products are built around models.
6away Engine is built around decisions.
…before trust.
ECOSYSTEM
…and future realities inside the 6away ecosystem.
ENTER THE ENGINE
A system for understanding, testing and deploying intelligence.
Built inside the 6away ecosystem. Designed to be reused outside it.