Karyalya · platform, governance & architecture

Autonomy you can put your name to.

Every agent is owned, measured, governed, and reversible. One policy engine at every gateway. A full audit trail on every action. This is the machinery under the muster — for the people who need to trust it before they sign.

See how it’s governed ↓ ← Back to the product tour

What makes a good agent.

Six qualities every agent on Karyalya carries — the same things you'd want in a good employee, plus the ones only software can offer.

Versioned & reversible

Every change to an agent is an immutable snapshot. Diff it, restore it, roll it back. There is no state you cannot return to.

draft → published → suspended → retired

It remembers

Four kinds of memory — contextual, episodic, semantic, strategic — scoped per user, per agent. It knows your business, not everyone's.

memory · scoped, expiring, private

It escalates

At the edge of its judgment, an agent stops and routes to the right human with full context — and a fallback if they're away.

escalation · pending → resolved / rerouted

It self-initiates

Watches for pressure across the business and proposes the work before you ask. Initiative, never unsupervised action.

self-initiation engine · proposals queued

It self-evolves

Learns from every task through evaluation and optimization — and only ships an improvement once you've approved the diff.

evaluation · optimization · your sign

It is governed

Bound by policy at every gateway, capped by budget, and logged without exception. Capability is never without a leash.

policy · budget · immutable audit

governance & policy · the Board Room

Governed like an enterprise, run like a small team.

One policy engine sits at every gateway. Write a rule in plain language or structured terms; Karyalya evaluates it in real time on every message, tool call, and workflow — before anything happens.

DENY Block the action outright — it never reaches the tool or the customer.

REQUIRE_APPROVAL Pause and route to a human. This is where f to ship lives.

PREPEND Inject a rule or disclaimer into the agent's context before it acts.

LOG Allow, but write it to the record for later review.

PII detection DLP Hallucination checks Script safety Credential hygiene Content safety

karyalya.app — Board Room

Product screenshot / GIF — the Board Room: policies, budgets, live guardrail hits, audit trail

100%

of agent actions written to an immutable, append-only log

1-click

evidence packet — date-range, event-type filtered, ready for a regulator

budget management

Spend caps per org, team, or agent. Warn at a threshold; block at the limit. Your AI workforce can never quietly run up a bill.

certifications & compliance

placeholder — drop verified certification badges

ISO 27001

in progress

SOC 2 Type II

in progress

HIPAA

ready

GDPR / DPDP

compliant

evaluation & optimization

Agents that get better — with a leash.

Karyalya reads every conversation for quiet signals — corrections, rephrasing, abandonment. An LLM-as-judge scores each agent, drafts a specific prompt change, and shows you the diff. It only ships on your approval. Turn on auto-optimize and it applies changes itself — but only above 80% confidence, and every change is still a versioned, reversible entry on the record.

Evolution you can read, approve, and undo.

step 1 / 6

Agents work

Every conversation runs across the Floor.

Agents work

Signals captured

Evaluation

Optimization drafted

You approve

New version ships

self-initiation engine · agent runtime

Good agents don't wait to be asked.

The runtime watches for pressure — a deadline slipping, a reconciliation drifting, a customer gone quiet — and proposes the work before you notice it's needed. It still doesn't act alone: proposals land in your queue, waiting for the mark.

01 · notices

Monitors drive pressures across your business — the things a diligent employee would keep half an eye on.

02 · proposes

Generates and prioritises task candidates — with the reasoning attached, so you know why it thinks this matters now.

03 · waits

Queues the proposal for a human. Initiative is welcome; unsupervised action is not. The mark is always yours.

system architecture · five-layer model

Built headless, on purpose.

No service holds UI logic. Every surface — the Floor, a partner's own app, a raw API client — talks to the same backbone. Event-driven where it can be, database-per-tenant always, every action audited by default.

L1 Client

Floor · Foundry · Board Room · external clients (API / SDK) — HTTPS + real-time to the gateway

L2 API Gateway

API Gateway · Auth · Rate Limiter · Request Router — authenticates and routes every inbound call

L3 Application Services

13 core microservices — sync over REST / gRPC, async over Kafka events

L4 Data

PostgreSQL (per-tenant) · Redis · Pinecone · S3 · HashiCorp Vault

L5 Infrastructure

Kubernetes · Temporal · Kafka cluster · multi-cloud — durability and lifecycle

Security & certifications

Built to pass the audit you’ll face.

Your data stays isolated to your own environment. Every consequential action carries a human signature and an immutable log entry. We build to the standards your enterprise customers and your auditor will ask about.

Data isolation per clientAppend-only audit trailRole-based accessHuman sign-off on high-risk actionsSOC 2 & ISO 27001 — on the roadmapDPDP-ready data handling

[ Update this list as certifications are achieved. Items marked “on the roadmap” are targets, not current attestations. ]