anusara 1.3.0

Deterministic execution engine with traceable, immutable history

Anusara

Deterministic execution runtime for agent workflows.

---

What is Anusara?

Anusara is a deterministic runtime for executing agent workflows.

It executes graphs of agents using:

• deterministic scheduling
• event-sourced execution history
• replay-driven state reconstruction

Every execution becomes:

• replayable
• inspectable
• debuggable
• reproducible

Anusara treats workflow execution as a runtime system problem, not a collection of loosely connected function calls.

---

🚀 Quickstart (2 minutes)

Install

pip install anusara[api]

Run the server

PYTHONPATH=src python -m uvicorn anusara.adapters.http.app:create_app --factory --reload

Register agent

curl -X POST http://127.0.0.1:8000/agents/register \
  -H "Content-Type: application/json" \
  -d '{"name": "hello"}'

Execute

curl -X POST http://127.0.0.1:8000/executions \
  -H "Content-Type: application/json" \
  -d '{"entry_nodes": ["hello"]}'

👉 You just ran your first deterministic workflow.

➡️ Full guide: docs/guides/http-api-quickstart.md

---

The Mental Model

Agent Graph
     ↓
Deterministic Execution Engine
     ↓
Event Log (source of truth)
     ↓
Replay • Debug • Observability

Execution is not hidden state.

It is a recorded, replayable system.

---

Why Anusara?

Most agent orchestration systems rely on implicit runtime behavior.

This leads to:

• nondeterministic execution
• difficult debugging
• hidden state transitions
• unreliable retries
• poor reproducibility

Anusara solves this by enforcing:

• deterministic execution
• event log authority
• replay-based debugging
• explicit lifecycle guarantees

---

Architecture Overview

flowchart LR

Client[Client / API / CLI]
Client --> Runtime

Runtime[Execution Runtime]

Runtime --> Graph[Graph Compiler]
Runtime --> Registry[Agent Registry]
Runtime --> Router[Router]
Runtime --> Mutation[Mutation Engine]

Runtime --> EventLog[(Event Log)]

EventLog --> Replay[Replay Engine]
EventLog --> Debugger[Debugger / Analysis]

Runtime --> Observability[Observability]

---

Core Responsibilities

Layer	Responsibility
Runtime	deterministic execution of workflows
Event Log	authoritative execution history
Registry	agent lifecycle & resolution
Router	execution decision logic
Mutation Engine	controlled graph evolution
Observability	debugging, metrics, visualization

---

Local Python Example

import asyncio
from anusara import ExecutionEngine, ExecutionRequest, AgentRegistry

engine = ExecutionEngine()

registry = AgentRegistry()
registry.register("hello", HelloAgent())

request = ExecutionRequest(
    request_id="demo",
    entry_nodes=["hello"],
    payload={},
    metadata={},
)

asyncio.run(engine.execute(request, registry))

---

Key Features

• deterministic execution
• event-sourced execution history
• replayable workflows
• HTTP + CLI + Python runtime
• agent lifecycle management
• built-in observability
• controlled workflow mutation

---

Documentation

🚀 Getting Started

• docs/guides/http-api-quickstart.md
• docs/guides/running-anusara-locally.md
• docs/guides/creating-your-first-agent.md

📚 Concepts

• agents
• deterministic execution
• event sourcing
• replay

🏗 Architecture

• execution model
• runtime architecture
• event log
• mutation model

📖 Reference

• HTTP API
• agent registry
• execution engine

---

Project Status

Anusara is under active development.

Current focus:

• developer experience (HTTP API & usability)
• onboarding and documentation
• production readiness

---

Contributing

Contributions and discussions are welcome.

• CONTRIBUTING.md
• CODE_OF_CONDUCT.md
• SECURITY.md

---

Security

Anusara Docker images are built on python:3.12-slim-bookworm and regularly updated to minimize vulnerabilities.

We prioritize:

• eliminating high/critical vulnerabilities
• maintaining minimal runtime footprint
• reproducible builds

Low-severity base image CVEs are tracked but not always actionable.

---

License

MIT License