Skip to content

Wayam Manthan · Architecture

How Manthan describes an agent

Three vocabularies, each with one job: the operating loop says where an agent works; the evidence tier says what Wayam can show for it; the architecture layers are the internal taxonomy the loop is folded from.

The Agent Operating Loop

Every Wayam solution is composed across four responsibilities. Manthan shows where each agent operates, what it may change, and where human approval is required.

  1. Ingest

    Connect the systems of record and read the signals the work already produces.

    21 patternsthen

  2. Reason

    Ground context, score options against policy and the stated goal, draft the next step.

    333 patternsthen

  3. Act

    Execute an approved step in the system of record and keep the evidence.

    374 patternsthen

  4. Govern

    Set policy, gate consequential actions on a named approver, and audit what ran.

    122 patterns
Human-in-the-loop gate: Consequential writes at the Act layer mandate named human sign-off with tamper-evident audit logging.

Evidence tiers

A tier describes evidence, not aspiration. Nothing moves above T4 without a named owner, a validation date and the artefacts the tier requires. Unreviewed claims expire back to T4.

TierLabelMinimum evidenceAllowed claims
T1
Production validatedLive customer workflow with approved proof, a production date, and a measured operating metric.Production claims, measured outcomes, actual stack where permitted.
T2
Pilot validatedCompleted customer or design-partner pilot with acceptance results and artefacts.Pilot results and bounded outcomes.
T3
Wayam demonstratedReproducible Wayam environment with a run trace, an evaluation set and a build version.Demonstrated capability and tested integrations.
T4
Reference patternCurated solution design with an owner and a review date. No performance or production claim.Possible workflow, typical stack, prerequisites.

Integration status

verified
Verified integration
tested
Tested connector
typical
Typical enterprise system
unknown
Not yet specified

Metric status

measured
Measured
pilot-observed
Pilot-observed
benchmark-derived
Benchmark-derived
scenario-only
Scenario only

Agrani (agrani — foremost, the one who leads) marks the thirty capabilities Wayam is prioritising for evidence. An Agrani badge is a commitment to attach a run replay, an evaluation set and a validation date; it is not itself proof.

Architecture view (L1 to L7)

The layer model inherited from the reference atlas. It is too granular and too unevenly populated to navigate by, so it is optional in the library. Every layer folds into one loop responsibility.

  • L1 folds into Govern

    Strategy

    Portfolio choices, goals and constraints for the estate.

  • L2 folds into Govern

    Operating model

    Ownership, approvers, and the operating cadence.

  • L3 folds into Govern

    Governance

    Policy, controls, evidence and audit.

  • L4 folds into Ingest

    Data foundation

    Connectors, signal capture and grounding data.

  • L5 folds into Reason

    Reasoning

    Research, scoring, planning and drafting.

  • L6 folds into Act

    Execution

    Approved write-back into systems of record.

  • L7 folds into Act

    Autonomous

    Bounded autonomous execution under standing policy.