Control-first automation

Build with accountable agents.

Policy-gated execution with clear provenance, approvals, and recovery.

Lifecycle at a glance
  1. Pre-Bloom
  2. Context Assembly
  3. Calyx Gate
  4. Approval
  5. Anthesis
  6. Dormancy

Project goals

Ensure autonomous actions only occur when context is complete, risk is understood, and authority is explicit.

Governed autonomy

Agents act within policy, approvals, and deterministic state transitions.

Human authority

Humans remain the final arbiters through review, override, and ownership.

Traceability

Every decision and transition is auditable and replayable.

Benefits

Operate AI with the same rigor you expect from production systems while maintaining flexibility.

Policy enforcement

Calyx policies define approval requirements and risk tiers.

Deterministic execution

Same inputs, same context, same policy, same outcome.

Audit-ready

Capture approvals, transitions, and evidence for compliance reviews.

SDLC overview

Each SDLC phase has a dedicated agent, with MCP coordinating handoffs and governance.

SDLC agents

Cross-functional teams

Operate in each phase they own, from product to QA.

Agents act through MCP

For all project interactions, enforcing policy and auditability.

Manual & Automated

Both human and AI-driven actions are governed by the same policies.

SDLC overview diagram

Control surfaces

CLI, UI, API

Tools for operators

Slack & Email approvals

Simple approval workflows.

Git PRs

Events that trigger workflows.

Direct repo edits

Manual changes trigger policy evaluation.

Usage scenarios

Two ways Anthesis operates in practice, with humans always in control.

Engineer-driven SDLC

Sequential, policy-gated runs led by humans.

  1. An engineer triggers agents in sequence for requirements, build, and testing.
  2. Each phase generates a proposal at risk thresholds, then resumes after approval.

Automated repo monitoring

Event-driven orchestration with approvals on demand.

  1. A repo event (commit or PR update) triggers a policy-evaluated workflow.
  2. MCP assembles context, classifies risk, and queues actions until approvals are granted.

Approvals and governance

Approvals are explicit, recorded, and tied to policy versions.

Approval surfaces

CLI, Slack, email, and Git-based review channels authorize proposals with traceable identities.

Policy gating

Calyx enforces risk tiers and required approvers.

Audit metadata

Decisions bind to commits, timestamps, policy versions, and scope.

Evidence trail

Logs, provenance, and trace IDs link decisions to outcomes.

Human-in-the-loop

Manual and autonomous workflows coexist without conflict.

Direct edits

Humans can edit the repository and trigger workflows via commits.

Overrides

Operators can override, halt, or replay any past execution.

Context curation

Approve or edit context artifacts and inputs.

Operational monitoring

Inspect logs, traces, and evidence before proceeding.

Transparent decisions

Every action is recorded with context and rationale.

Policy authoring

Define Calyx rules, risk tiers, and required approvers.

Collaborative reviews

Multiple approvers can review and sign off on proposals.

Post-hoc accountability

Audits, incident reviews, and retrospective sign-off.

System pillars

Three pillars define how Anthesis executes, interacts, and assures outcomes.

Execution core

Phloem MCP server

Orchestration + state machine.

Xylem dispatch fabric

Async execution + routing.

Automation + models

Runtime abstraction + workflows.

Interaction

Git repository

System of record.

Control surfaces

CLI / UI / API / chat.

Inflorescence

Multi-agent sequencing and handoffs.

Assurance

Observability

Logs, traces, and evidence export.

Provenance

Attribution for human and agent artifacts.

Audit trail

Immutable decision history.

Lifecycle

Every autonomous action follows a governed, auditable state machine.

Pre-Bloom

Record the triggering event, register the execution, and validate prerequisites.

Context Assembly

Gather RFCs, requirements, and artifacts, then bind them into a deterministic context snapshot.

Calyx Gate

Evaluate policy and Bloom Class risk, determine approval requirements, and block unsafe paths.

Approval

Capture explicit authorization, record audit metadata, and bind the decision to outputs.

Anthesis

Execute with full context, apply governed actions, and log structured outcomes.

Dormancy

Complete, roll back, or halt safely, preserving traceable outcomes and evidence.

Contributing and contact

Anthesis is under active development with RFC-driven architecture. Contributions are welcome.

Contribute

Open an issue or pull request in the private repository to propose improvements.

Contact

Reach out to the maintainers through Micrantha.

Governance

Review the RFCs in meristem/ to understand the living architecture.

FAQ

Common questions about Anthesis.

What is Anthesis?

Anthesis is a governed agentic SDLC system that ensures autonomous actions only occur when context is complete, risk is understood, and authority is explicit. By enforcing policy, approvals, and traceable state transitions, Anthesis enables teams to leverage AI-driven agents while still maintaining human work, oversight and accountability.

Can Anthesis support multi-repo projects?

Yes. MCP can coordinate events and approvals across repositories, while Git remains the source of truth for each repo.

Can SDLC phases run in parallel?

Yes. Anthesis supports parallel agent workflows with explicit coordination and policy gates to prevent unsafe overlap.