Documentation · Architecture

How a signal becomes evidence-backed root cause.

Diagnara is a governed, tool-driven, multi-agent platform. Every input is scoped by the control plane, orchestrated through investigation runbooks, sourced only from registered tools, and traced back to the external systems where the evidence actually lives — flowing into one auditable evidence store.

Governed by defaultTool-first evidenceHuman-in-the-loopMulti-agent orchestration

System map

Components & external communication

Left to right: a signal enters, passes the governance gate, is investigated by orchestrated agents, which reach out through governed tools to external systems — every fact returns to the evidence store, then becomes a defensible conclusion and reusable knowledge. Hover any node to trace its connections.

Intake & controlOrchestrationTool accessExternal & evidence

Scroll horizontally to explore the full map

01 Signals

Observability alert

p95 · error rate · saturation

Jira / ServiceNow

change & incident ticket

Manual prompt

engineer-initiated

02 Control plane

Governance gate

enforced before anything runs

RBAC & environment scope

Policies, quotas & approvals

Model routing & cost controls

Immutable audit trail

03 Investigation layer

Investigation runbook

orchestrator · versioned

Incident Investigator

agent

SRE Assistant

agent

Contract Analyst

agent

Runbook Executor

agent

04 Governed tools

Logs

rest

Metrics & APM

traces

Deployments & PRs

deploy events

Tickets

itsm

SQL & APIs

rest · graphql

Billing & contracts

rest api

05 External systems

Elastic

log store

Datadog

metrics · apm

GitHub

source · prs · deploys

Jira · ServiceNow

tickets

Postgres

prod · replica

Billing API

ledger · contracts

Evidence store & timeline

Every tool call returns a cited, source-linked data point — accumulated into one reproducible, auditable record.

cited evidencetimelineaudit events

ConcludeEvidence-backed report

Ranked root cause, confidence label, impact assessment & next actions — every claim source-linked, every sensitive call gated by human approval.

InstitutionalizeReusable knowledge

Timeline, evidence graph & report are stored as auditable knowledge — so the next similar incident resolves far faster.

Agents never invent facts — they may only act through registered tools, within approved environments.