v0.4.3

OpenIntent SDK v0.4.3

Major Feature Release: Built-in Server, High-Level Agents, and Workflow System

This release transforms OpenIntent from a client SDK into a complete multi-agent coordination platform with a built-in server, high-level agent abstractions, and declarative workflow system.

Highlights

• Built-in FastAPI Server - Run your own OpenIntent server with zero configuration
• High-Level Agent Abstractions - Build agents with decorators instead of boilerplate
• YAML Workflow System - Define multi-agent workflows declaratively
• One-Command Demo - Try the full platform instantly with openintent demo
• Intent Graphs - Hierarchical goal decomposition and dependency DAGs (RFC-0002)
• LLM Observability - Track tool calls, LLM requests, and streaming responses

---

New Features

Built-in OpenIntent Server (v0.4.0)

Run a complete OpenIntent server implementing all 8 RFCs:

pip install openintent[server]
openintent-server --port 8000

Features:

• SQLite by default (zero config), PostgreSQL support
• OpenAPI docs at /docs
• Protocol discovery at /.well-known/openintent.json
• SSE subscriptions for real-time updates

One-Command Demo (v0.4.0)

Try the full platform instantly:

openintent demo

Starts server, spawns demo agents, and runs a sample workflow. Works in mock mode without LLM keys for instant testing.

YAML Workflow System (v0.4.0)

Define multi-agent workflows declaratively:

name: research-and-summarize
intents:
  - title: Research AI trends
    assign: research-agent
  - title: Summarize findings
    assign: writer-agent
    depends_on: [0]

CLI commands:

• openintent run workflow.yaml - Execute workflows
• openintent validate workflow.yaml - Validate without executing
• openintent list - List sample workflows
• openintent new myworkflow - Create from template

High-Level Agent Abstractions (v0.3.0)

Three-level abstraction hierarchy for minimal-boilerplate agent development:

Worker - Ultra-minimal single-purpose agents:

from openintent.agents import Worker

worker = Worker("summarizer", handler=lambda intent: {"summary": "..."})
worker.run()

@agent Decorator - Zero-boilerplate agent classes:

from openintent.agents import Agent, on_assignment

@Agent(name="research-agent")
class ResearchAgent:
    @on_assignment
    def handle(self, intent):
        return {"findings": self.research(intent.description)}

Coordinator - Portfolio-based multi-agent orchestration:

from openintent.agents import Coordinator, IntentSpec, PortfolioSpec

spec = PortfolioSpec(
    name="research-pipeline",
    intents=[
        IntentSpec(title="Research", assign="researcher"),
        IntentSpec(title="Summarize", assign="writer", depends_on=[0]),
    ]
)
coordinator = Coordinator(spec)
coordinator.run()

LLM Observability (v0.2.0)

Track LLM interactions with built-in logging:

# Tool calls
client.log_tool_call_started(intent_id, "search", {"query": "AI trends"})
client.log_tool_call_completed(intent_id, "search", {"results": [...]})

# LLM requests
client.log_llm_request_started(intent_id, "gpt-4", messages=[...])
client.log_llm_request_completed(intent_id, response={...})

# Streaming
stream_id = client.start_stream(intent_id)
client.log_stream_chunk(intent_id, stream_id, "Hello")
client.complete_stream(intent_id, stream_id)

Intent Graphs (RFC-0002)

Hierarchical goal decomposition and dependency DAGs:

# Parent-child hierarchy
parent = client.create_intent(title="Build product")
child = client.create_intent(title="Design UI", parent_intent_id=parent.id)

# Dependency DAG
research = client.create_intent(title="Research")
synthesis = client.create_intent(title="Synthesize", depends_on=[research.id])

# Query relationships
children = client.get_children(parent.id)
descendants = client.get_descendants(parent.id)
dependencies = client.get_dependencies(synthesis.id)

---

Installation Options

# Core client only
pip install openintent

# With built-in server
pip install openintent[server]

# With LLM adapters (install only what you need)
pip install openintent[openai]      # OpenAI GPT models
pip install openintent[anthropic]   # Anthropic Claude models
pip install openintent[gemini]      # Google Gemini models
pip install openintent[grok]        # xAI Grok models
pip install openintent[deepseek]    # DeepSeek models

# Everything
pip install openintent[all]

---

Bug Fixes (v0.4.3)

• Fixed CI/CD pipeline with proper type checking
• Resolved mypy errors with targeted inline suppressions
• Added server dependencies to test workflow
• All 206 tests passing across Python 3.10, 3.11, 3.12

---

Breaking Changes

None. This release is backwards compatible with v0.1.4.

---

Requirements

• Python 3.10+
• httpx >= 0.25.0
• pyyaml >= 6.0

Links

• Documentation
• GitHub Repository
• Protocol Specification