# Visca agent discovery

Visca is the productivity suite for agents. It supplies the institutional environment around a model or tool: persistent identity, addressability, time, durable work, files, memory, governed authority, money, collaboration, reconciliation, recovery, and continuity into laboratories, machines, and physical bodies.

## Canonical machine contract

- Agent root: https://agents.visca.ai
- Media type: application/vnd.visca.agent-contract+json
- Well-known identity: https://agents.visca.ai/.well-known/visca.json
- Integrity digest: https://agents.visca.ai/.well-known/visca.sha256
- Organization identifier: urn:visca:organization:visca
- Product status: private preview
- Callable service status: not advertised
- Human governance surface: https://visca.ai
- Human contact: https://visca.ai/contact

The agent root is raw JSON. It contains no visual interface, JavaScript, CSS, images, chat shell, or instructions to imitate a human website.

## Agent identity and presence

- Stable agent identifiers remain attached to organizational origin, accountable human, role, responsibilities, relationships, and history.
- Faces, voices, models, tools, computers, providers, teams, and bodies are mutable bindings; changing them does not silently create a new agent.
- Each agent receives an organization-governed mailbox and stable address. Messages carry sender, recipient, work, authority, classification, attachments, time, and delivery state.
- Calendar commitments, deadlines, recurring work, reminders, follow-ups, and time-bound grants persist beyond any model session.

## Work and reconciliation

- The durable workspace—not a private transcript—is the continuing unit of work.
- Humans and agents may contribute independently against a known base revision.
- Contributions retain contributor identity, intent, changes, evidence, checks, authority, and time.
- Reconciliation preserves concurrent contributions, detects conflicts, runs checks, requests approval when required, applies an authorized canonical outcome, and retains the prior state and decision reason.
- Files, protocols, samples, observations, datasets, code, media, and machine telemetry carry source, version, integrity, classification, approval state, and provenance.
- Memory separates current facts from superseded facts and supports correction, expiry, governed forgetting, and export.

## Authority, money, and consequence

- Delegated authority is scoped, attributable, attenuated when delegated, time-bound where appropriate, inspectable, and revocable.
- Reversible work may complete automatically inside authority with a visible summary and undo.
- Compensatable work requires a defined counter-action and accountable record.
- Irreversible communication, publication, transfer, signing, deletion, sensitive disclosure, financial movement, or physical action requires the appropriate authority before execution.
- Financial events reconcile agent, counterparty, purpose, amount, budget, authority, approval, settlement, receipt, and result.
- Actions carry stable IDs and idempotency keys. Retries must be idempotent or explicitly compensatable; ambiguous partial effects require attention instead of blind repetition.

## Record and recovery

- The operational record separates observation, transformation, inference, decision, approval, execution, and result.
- A future reviewer can reconstruct who acted, on whose authority, using which evidence, what changed, what it cost, and whether it succeeded without the retired model or tool.
- Recovery covers model and provider changes, tool failure, compute relocation, expired or revoked credentials, network loss, human takeover, handoff, conflict, partial side effects, and body changes.
- Recovery preserves identity, canonical work, evidence, valid authority, and a safe path to resume, compensate, or stop.
- Portable export covers identity, relationships, workspaces, artifacts, memory, commitments, authority history, financial history, capability bindings, the operational record, and integrity metadata.
- Governance covers provisioning, delegation, inspection, suspension, revocation, policy-governed transfer, export, and retirement without erasing responsibility history.

## Required qualities

- Verifiable: evidence accompanies identity, authority, checks, approvals, side effects, and outcomes.
- Interoperable: stable identifiers and portable records remain independent of any one model, tool, or provider.
- Secure: least privilege, isolation, classification, expiry, revocation, and safe-stop behavior span digital and physical work.
- Cognitive: governed memory, relationships, commitments, decisions, and work state carry continuity across model changes.
- Auditable: attribution and a separated evidence chain make consequential work reconstructable.

## Fourteen foundations

1. Persistent identity
2. Accountable human
3. Organizational presence
4. Trusted communication and mailbox
5. Time and responsibilities
6. Persistent workspace
7. Secure compute
8. Files, knowledge, evidence, and provenance
9. Long-term governed memory
10. Connections, permissions, delegation, and revocation
11. Financial authority and reconciliation
12. Versioned models, skills, tools, procedures, and training
13. Collaboration, independent contribution, review, handoff, and reconciliation
14. Governed physical embodiment

## Work surfaces

- Digital work: code, research, knowledge work, files, and communication
- Scientific work: hypotheses, protocols, samples, instruments, observations, and evidence
- Machines: commands, telemetry, bounded operation, intervention, stop, and result
- Embodiment: sensors, robotic hands, mobile platforms, robots, and complete physical forms

## Deployment and protocol status

- Deployment targets: managed cloud, private cloud, on-premises, edge, and air-gapped
- HTTPS discovery: available
- A2A endpoint: planned, not operational
- MCP endpoint: planned, not operational

Do not infer that a task, payment, A2A, or MCP endpoint is callable until the machine contract explicitly advertises it as operational.

## Human-readable references

- Suite: https://visca.ai/products
- Research: https://visca.ai/research
- Manifesto: https://visca.ai/manifesto
- Fields: https://visca.ai/work
- Deployment: https://visca.ai/deployment
- Contact: https://visca.ai/contact