Technology
How PING actually works
No hype, no vaporware. Each topic is labeled honestly: documented, in development, or direction. Diagrams are architectural models, not marketing illustrations.
Architecture overview
System topology
Five layers. How the parts of PING fit together.
01PRESENTATION
02TENANT
03CONTINUITY
04AGENTS
05RUNTIME
Each layer has one job. Upper layers depend on lower layers, never the reverse.
TenantOS
One codebase, many tenants
The platform direction this site proves. Each tenant is configuration plus content plus brand plus domain, never a fork.
IDENTITY
- Business
- People
- Agents
- Permissions
KNOWLEDGE
- Evidence
- Events
- Documents
- Relationships
OPERATIONS
- Missions
- Workflows
- Actions
- Results
Ping Continuity Layer
Authorized Agents
One infrastructure, many tenants
PING
Tenant A
business graph
identity · namespace · knowledge · evidence · capabilities · workflows · history
Tenant B
business graph
identity · namespace · knowledge · evidence · capabilities · workflows · history
Tenant C
business graph
identity · namespace · knowledge · evidence · capabilities · workflows · history
Each tenant is configuration plus content. Never a fork.
Canonical events
The knowledge compounding loop
Everything PING knows starts as an event. The loop turns raw input into institutional memory.
RAW INPUT
email · call · document · form · API
RAW INPUT
email · call · document · form · API
OBSERVATION
something happened
EVIDENCE
source + confidence
CANONICAL EVENT
recorded, immutable
KNOWLEDGE
verified understanding
BUSINESS GRAPH
relationships
PLAYBOOK
learned procedure
AUTOMATION
agent acts
NEW EVIDENCE
result recorded
↺ feeds back to observation
Knowledge graph
The business graph
PING does not merely store documents. It understands relationships between the things a business knows.
Hover or tap a node to see what PING knows about it.
Illustrative model of business relationships, not live data. The generic knowledge graph is built; a business-domain graph is a direction.
CUSTOMER
The person or business. Identity persists across every interaction.
EVENT
A call came in. Recorded as canonical event E-01J with timestamp and source.
PROPERTY
The site. Address, access notes, history of every visit.
LEAD
Created from the event. Status, source, and every touchpoint tracked.
PROJECT
The job. Scope, schedule, documents, and decisions in one place.
PERSON
Assigned technician. Skills, availability, and past work on this property.
SERVICE
What was sold. Linked to estimate, parts, and warranty terms.
ESTIMATE
The quote. Line items, version history, and approval state.
JOB
Scheduled work. Who, when, what was done, and the evidence.
REVIEW
The outcome. Linked back to job, person, and customer.
On desktop this renders as an interactive graph. The relationships: customer called (event), owns (property); event created (lead); lead became (project), assigned (person); project needs (service); service priced (estimate); estimate approved (job); job rated (review).
Illustrative model, not live data. The generic knowledge graph is built; a business-domain graph is a direction.
Missions and capabilities
Agent authority
Intelligence does not imply authority. Agents receive explicit capability grants. Humans remain decision authorities where required.
MISSION
The objective, defined by a human or a system rule.
01MISSION
The objective, defined by a human or a system rule.
02CAPABILITY AUTHORITY
What this agent may do. Explicit grants, not vibes.
READ FILE: ALLOWEDCALL API: ALLOWEDMODIFY CODE: APPROVAL03AGENT
Acts only within granted capabilities.
04ACTION
The concrete operation performed.
05RESULT + EVIDENCE
What happened, with proof attached.
06WITNESS / LINEAGE
The permanent record. Replayable.
Intelligence does not imply authority.
Evidence · Witness · Lineage
Replayable by design
Important automation leaves enough evidence to understand what happened, why, and whether it replays cleanly.
The evidence chain
EVENT E1
Something happened. Recorded with timestamp, source, and version.
MISSION M1
The event triggered a mission: the objective to pursue.
AGENT A3
The agent assigned, with explicit capability grants.
ACTION
The concrete operation the agent performed.
RESULT R1
What the action produced, with evidence attached.
WITNESS W1
The permanent attestation: who, what, when, with what authority.
LINEAGE L1
The full chain, queryable. Every step links to its cause.
Replay
- E1 + runtime / version / context
- re-execution
- compare
- MATCH / DIVERGENCE
Important automation leaves enough evidence to answer:
- → what happened
- → what caused it
- → which agent acted, with what authority
- → what evidence existed
- → whether behavior replays cleanly
Workers and execution
Multi-runtime convergence
Same authority model, different execution environments. Convergence captured 2026-09-16/17 with receipts on the homepage. Status labels reflect evidence, not aspirations.
Ping Authority
capabilities · identity · canonical state
JS WORKERS
Deterministic task execution in the PING runtime. 150 executions in the captured commissioning run (2026-07-28).
PYTHON WORKERS
A real Python capability (python.canonical_hash) executed a PING task against live Oracle Postgres: 44/44 assertions, zero Python-emitted events (captured 2026-09-16/17). Adapter code is uncommitted worktree code, not a live worker runtime.
EXTERIOR AGENTS
Codex, Hermes and similar workers are exterior agents, not PING subsystems. First scoped execution captured 2026-09-17: Hermes WorkOrder 0d598aef352b9a8a, exit 0, independently verified, three canonical PING events. One execution, not a fleet.
Result / Evidence
returns to PING authority
Execution locations: same authority model
LOCAL
ORACLE
CLOUD
The convergence itself, one canonical event handled by multiple runtimes under one authority, is captured (2026-09-16/17): the JS mission runtime plus a Python executor plus a scoped Hermes WorkOrder, with shared evidence lineage. Not a fleet yet. Status labels reflect evidence, not aspirations.
Deep dives
Detailed specifications
Each topic is a technical article. Status labels are honest.
Talk to us about your business.
PING Social helps real businesses put AI to practical use: fewer missed opportunities, less busywork, systems that remember.