v0.2.0 — Developer Experience Release

What's New in v0.2.0

This release focuses on developer experience — making ModelMesh easier to adopt, test, debug, and extend. All features ship with full Python + TypeScript parity.

7 New Features

1. Structured Exception Hierarchy

Catch specific error types instead of bare RuntimeError:

from modelmesh.exceptions import RateLimitError, BudgetExceededError

try:
    response = client.chat.completions.create(...)
except RateLimitError as e:
    print(f"Retry in {e.retry_after}s")
except BudgetExceededError as e:
    print(f"{e.limit_type} limit: ${e.limit_value}")

2. Request/Response Middleware

Intercept every request for logging, transforms, or custom headers:

from modelmesh import Middleware

class LoggingMiddleware(Middleware):
    async def before_request(self, request, context):
        print(f"-> {context.model_id}")
        return request
    async def after_response(self, response, context):
        print(f"<- {response.usage.total_tokens} tokens")
        return response

3. Context Manager Support

Clean resource cleanup with with/async with (Python) or close() (TypeScript):

async with modelmesh.create("chat") as client:
    response = await client.chat.completions.create(...)
# Automatic cleanup

4. Usage Tracking API

Monitor costs and tokens across models and providers:

print(client.usage.total_cost)
print(client.usage.by_model)

5. Mock Testing Client

Test without live APIs — pre-configured responses with call recording:

from modelmesh.testing import mock_client, MockResponse
client = mock_client(responses=[MockResponse(content="Hello!", tokens=10)])

6. Capability Discovery

Explore available capabilities without memorizing paths:

modelmesh.capabilities.resolve("chat-completion")
# -> 'generation.text-generation.chat-completion'
modelmesh.capabilities.search("text")
# -> ['text-embeddings', 'text-generation', ...]

7. Routing Explanation

Debug why a model was selected without making API calls:

explanation = client.explain(model="chat")
print(explanation.selected_model, explanation.reason)

Audit & Quality

• 50+ code-docs-tests inconsistencies fixed
• Cross-language parity audit: matching method signatures, parameters, return types
• All new features backward-compatible — existing code works unchanged

Testing

• 1,879 tests passing (1,166 Python + 713 TypeScript)
• 63 new Python DX tests + 69 new TypeScript DX tests

Samples & Documentation

• 12 new quickstart samples (06-11 for Python and TypeScript)
• 5 new developer guides: QuickStart, Error Handling, Middleware, Testing, Capabilities
• Updated docs index with new features and navigation

Install

# Python
pip install modelmesh-lite==0.2.0

# TypeScript
npm install @nistrapa/modelmesh-core@0.2.0

TypeScript Sub-path Imports

New deep imports for tree-shaking:

import { mockClient } from '@nistrapa/modelmesh-core/testing';
import { resolve } from '@nistrapa/modelmesh-core/capabilities';
import { Middleware } from '@nistrapa/modelmesh-core/middleware';

Full Changelog: v0.1.1...v0.2.0

v0.1.1 — Package Publish & Docs Update

What's New

Published Packages

• PyPI: pip install modelmesh-lite
• npm: npm install @nistrapa/modelmesh-core
• Docker: ghcr.io/apartsinprojects/modelmesh:v0.1.1

Changes since v0.1.0

Package Rename

• npm package renamed from @modelmesh/core to @nistrapa/modelmesh-core
• Updated all 50 files: docs, samples, scripts, and package.json

New Documentation

• AI Agent Integration Guide (docs/ForAIAgent.md) — decision tree, all 4 integration paths, YAML config reference, troubleshooting
• Claude Skills (ClaudeSkill/) — 5 ready-to-use skills for install, configure, integrate, deploy, and test

Fixes

• Fixed publish workflow secret name (PYPI_TOKEN → PYPI_API_TOKEN)
• Fixed TypeScript build errors (RUNTIME static property type widening on CDK base classes)
• Added py.typed marker (PEP 561) for mypy/pyright support
• Added exports field to TypeScript package.json for subpath imports

Infrastructure

• Docker proxy deployment with OpenAI-compatible REST API
• CI/CD publish workflow for PyPI, npm, and GHCR
• 1,366 tests passing (855 Python + 511 TypeScript)

Install

# Python
pip install modelmesh-lite

# TypeScript
npm install @nistrapa/modelmesh-core

# Docker
docker pull ghcr.io/apartsinprojects/modelmesh:v0.1.1

Quick Start

import modelmesh
client = modelmesh.create("chat-completion")
response = client.chat.completions.create(
    model="chat-completion",
    messages=[{"role": "user", "content": "Hello!"}],
)

v0.1.0 - Initial Release

ModelMesh Lite v0.1.0

First release of ModelMesh Lite -- capability-driven AI model routing.

Highlights

• Core: Router, capability tree, pools, state management, event emitter
• CDK: Base classes for all 6 connector types with mixins (cache, metrics, rate limiter, HTTP client)
• Pre-shipped connectors: OpenAI and Anthropic providers
• MeshClient: OpenAI SDK-compatible client (chat.completions.create)
• Config: YAML loader, auto-detect for 9 providers
• Observability: Routing events, request logging, aggregate statistics
• 356 tests across 11 test modules
• Zero external dependencies for core

Install

pip install modelmesh-lite

Or install the wheel directly:

pip install modelmesh_lite-0.1.0-py3-none-any.whl

Quick Start

import modelmesh

client = modelmesh.create("chat-completion")
response = client.chat.completions.create(
    model="chat-completion",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)