The Formation Roadmap

Transitioning an organization to the agentic standard requires structured gates and protocol compliance. The Niyatna standard defines a 5-phase formation pathway to establish, run, and scale an Agentic Company.

---

Phase 1: Intent Mapping & Drag Audits

Before deploying agent runtimes, founders must audit operational drag and explicitly document the company's intent.

• Drag Auditing: Identify repetitive administrative or analytical processes draining 10–40+ hours per week (e.g., payment reconciliation, customer provisioning, manual reporting).
• Intent Specification: Translate business rules, data requirements, and taste preferences into structured parameters.
• Standard Boundary Design: Define what must be automated, what requires human validation, and what must remain blocked.

Phase 2: Agent Architecture Design

With the intent mapped, founders design the specialized agent workforce configuration. Roles are designed to map directly to operational departments:

• Niyatna CEO Agent (Strategic Catalyst): Monitors high-level operational indicators, runs resource models, and prepares executive briefings.
• Niyatna COO Agent (Execution Engine): Manages data pipelines, webhooks, customer onboarding, and coordinate notifications.
• Niyatna CFO Agent (Financial Guardrail): Audits ledgers, performs transaction matching, and flags margin discrepancies.

Each agent is provisioned with a dedicated mission, goal set, tools, and visual identity.

Phase 3: Runtime Provisioning (Niyatna OS)

Runtimes are provisioned in secure, isolated sandboxes. The Niyatna OS layer handles infrastructure setup:

• Sandbox Isolation: Spawns jailed virtual runtimes (isolated agent containers) restricting system path access.
• Credential Routing: Integrates API keys and certificates using LocalRoute, delivering scoped session tokens to runtimes without exposing master secrets.
• Command Boundaries: Disables shell write access and external webhooks unless explicitly whitelisted in the workspace config.

Phase 4: Intent Delegation (OpenIntent)

Daily operations are dispatched using OpenIntent packets. Rather than vague, conversational chat prompts, tasks are structured into strict schemas containing:

1. Goals: The target outcomes.
2. Context: Pointers to local memory, schemas, and live database scopes.
3. Constraints: Budget caps, time limits, and disallowed libraries.
4. Proof Standards: Defined evidence (e.g., git diffs, payment logs, transaction receipts) required for verification.

Phase 5: Verification & Veto (Proof of Intent)

Once execution completes, Niyatna OS parses the output and returns a structured Proof of Intent bundle via Niyatna HQ:

• Evidence Inspection: Displays execution logs, file diffs, and API receipts.
• Consensus Verification: Compares agent output against validation rules.
• Human Veto: The operator reviews the proof and either approves the changes to the production system or triggers a rollback.
• Strategic Leverage: Completed tasks return time back to the founders, reclaiming space for high-level decision making.