Skip to main content
Connic
Back to BlogProduct Spotlight

Composer SDK: Better Agent Development Tooling

Stop manual uploads and YAML guessing. The Composer SDK adds scaffolding, validation, cloud-backed hot-reload development, and CLI deployments.

December 27, 2025(last updated: July 19, 2026)5 min readAuthor: Connic Engineering

We have always supported defining agents in YAML and writing tools in Python. Until now, setup and deployment still involved considerable manual work: uploading files, creating project structures, and discovering YAML typos after the fact. Today we're releasing the Composer SDK to change that.

What Was the Problem?

Building agents on Connic worked, but the developer experience had friction:

  • •Manual uploads: Every agent or tool change required another file upload
  • •No scaffolding: Developers had to create the folder structure and boilerplate files from scratch
  • •No local validation: YAML syntax errors or missing tools appeared only after uploading
  • •Slow iteration: The upload-test-fix cycle was tedious, especially for small changes

The SDK solves each of these with proper CLI tooling.

Build a first agent with Composer

Agents stay in YAML, tools remain ordinary Python functions, and a push to Git starts deployment.

Try Connic free

What the SDK Does

Install it from PyPI:

Terminal
$ pip install connic-composer-sdk

The package provides commands for the full workflow:

connic init - Project Scaffolding

No more setting up folders by hand. One command creates a complete project structure with example agents, tools, middleware, and a README explaining each piece.

Terminal
$ connic init my-agents

āœ“ Initialized Connic project in /Users/you/my-agents

Created files:
  agents/assistant.yaml          (LLM agent)
  agents/invoice-processor.yaml  (LLM agent with retry)
  agents/tax-calculator.yaml     (Tool agent)
  agents/document-pipeline.yaml  (Sequential agent)
  agents/orchestrator.yaml       (Orchestrator with trigger_agent)
  agents/retrieval-agent.yaml    (Retrieval agent with RAG)
  tools/calculator.py
  middleware/assistant.py
  .gitignore
  requirements.txt
  README.md

The generated examples cover all agent types (LLM, Tool, Sequential) plus patterns like orchestration and semantic retrieval. Unneeded examples can be deleted and the rest modified.

connic lint - Local Validation

Catch errors before uploading. The SDK loads all project agents and tools locally, validates the YAML, checks that referenced tools exist, and previews the deployment.

Terminal
$ connic lint

Connic Composer SDK - Validation

Discovering tools...
  Found 3 tools in 1 modules:
    calculator: add, multiply, calculate_tax

Discovering middlewares...
  Found middlewares for 1 agents:
    assistant: before, after

Loading agents...
  Found 3 agents:

  ā”Œā”€ assistant [LLM]
  │  Description: A helpful assistant with calculator tools
  │  Model: gemini/gemini-2.5-flash
  │  Tools: calculator.add, calculator.multiply, calculator.calculate_tax
  └─

  ā”Œā”€ tax-calculator [Tool]
  │  Description: Calculates tax directly
  │  Tool: calculator.calculate_tax
  └─

  ā”Œā”€ document-pipeline [Sequential]
  │  Description: Processes documents through extraction and validation
  │  Chain: assistant → invoice-processor
  └─

āœ“ Project validation complete

A misspelled tool reference, invalid YAML, or missing middleware appears immediately instead of after uploading.

connic dev - Hot Reload Development

A development session syncs local files to an isolated cloud runner. After a file is changed and saved, it is live in 2-5 seconds without a manual upload.

Terminal
$ connic dev

Watching for file changes...
  Press Ctrl+C to stop

────────────────────────────────────────────────────────────
  View and trigger your agents: https://connic.co/projects/...
────────────────────────────────────────────────────────────

[14:32:15] Detected change: assistant.yaml
[14:32:16] Files changed, uploading...
[14:32:17] Uploaded 2,847 bytes

The CLI watches agents/, tools/, middleware/, and requirements.txt. Connector tests in the dashboard reflect code changes within seconds.

connic deploy - CLI Deployment

Deploy from a terminal or CI/CD pipeline without going through the dashboard.

Terminal
$ connic login              # Save credentials
$ connic deploy             # Deploy to default environment
$ connic deploy --env <id>  # Deploy to specific environment

With Git integration, a repository push deploys automatically. The normal workflow does not require the CLI.

New: Import Predefined Tools

Alongside the SDK, predefined tools can now be imported into custom tools. This supports building on top of capabilities like semantic retrieval, web search, or agent orchestration directly from Python code.

tools/my_tools.py
from connic.tools import retrieval_query, retrieval_store

async def lookup_and_summarize(query: str) -> str:
    """Search the retrieval and return a summary."""
    # Use the predefined retrieval tool
    results = await retrieval_query(
        query=query,
        namespace="documents",
        top_k=5
    )

    # Process and return the results
    if not results:
        return "No relevant information found."

    return "\n".join([r["content"] for r in results])

Previously, predefined tools could only be used directly by agents. They can now be composed into richer tool logic: query retrieval, process the results, make decisions, return exactly the required output.

Get Started

The SDK is available now. Existing Connic YAML configurations and Python tools continue to work unchanged, with the new tooling around them.

Terminal
$ pip install connic-composer-sdk
$ connic init my-agents  # Or use with your existing project
$ cd my-agents
$ connic lint            # Validate everything locally
$ connic dev             # Start hot-reload development

The quickstart guide provides a complete walkthrough, and the SDK documentation contains the full reference.

More from the Blog

Product Spotlight

AI Agent Browser Automation: How It Works and How to Build It

Learn how AI agent browser automation works, when to use it, and how Connic adds managed browsing, persistent state, tabs, dialogs, and files.

September 21, 20268 min read
Product Spotlight

Give Your AI Agents Hundreds of Tools Without Overloading Their Context

Connic keeps a large tool catalog outside the model context and retrieves schemas on demand, so agents can use more tools without loading them all upfront.

September 14, 20267 min read
Product Spotlight

AI Agent Routing: Trigger Agents and Return Results

AI agent routing moves events into an agent and returns results to the right system through Connic sync routes and asynchronous outbound connectors.

August 24, 20269 min read
Product Spotlight

Staging to Production: How Connic Environments Isolate AI Agents

Connic environments map git branches to isolated deployments, each with its own secrets, connectors, budgets, and run history, so one agent spec ships safely.

August 6, 20268 min read
Product Spotlight

LLM Context Compression for Long-Running AI Agents

Connic compresses older conversation history and oversized tool results, then retries the model call, so long-running agent sessions survive context limits.

July 20, 20268 min read
Product Spotlight

A/B Testing for AI Agents: Ship Better Prompts with Confidence

A changed prompt may feel better. A controlled experiment shows whether it is and lets real traffic decide.

March 27, 20269 min read
Product Spotlight

Secure AI Agents: A Production Safety Checklist

A production security checklist for AI agents, updated with ENISA’s 2026 evidence: scoped tools, approvals, logging, and incident response for critical sectors.

March 21, 202616 min read
Product Spotlight

Connic Bridge: AI Agents for Private Infrastructure

Connic Bridge creates a secure outbound tunnel so AI agents can reach private Kafka, databases, and internal services without opening inbound ports.

February 19, 20267 min read
Product Spotlight

Agent Observability: Track Costs, Tokens & Runs

Deploying AI agents without visibility is flying blind. Build custom dashboards, track LLM costs per model, and catch failures before users do.

January 23, 20268 min read