FastMCP low-key turns building an MCP server, and hooking it into Claude Desktop, into a five-minute task.
Most of my DSPy Newsletter content discovery and filtering exercise before curating is in Claude Desktop via local MCP server calling API which in turns hits the DB.
Here are small self contained code snippets on how you can build one.
The memory implementation of nanobot codebase is deviously simple
Instructions to manage memory is via skills
MEMORY(dot)md - User persona and preferences and code project context. Injected with system prompts (context(dot)py loads it)
HISTORY(dot)md - Append-only log where summaries reside and are timestamped. ( LLM asked to use grep )
Skills from system prompt standpoint are split into mandatory and always injected, and for rest non-mandatory we have one liner summary injected. System prompt has instructions on where to load them from if need be (directory) and how to load them.
On memory front the system prompt is telling the model that you decide what and when to store in memory and use grep to retrieve. The Memory Manager implementation ( 138 loc ) does CUR of CURD on a file ( no delete but consolidate ).
On a high level we are saying
a) In Model we trust to iterate and self-discover the right answer/path forward,
b) skills(dot)md is the equivalent of Neo in Matrix picking whatever skills he needs on fly when need be.
This project reminds me of The Minix Book for teaching operating system by Andrew Tanenbaum. This project is a great learning tool for seeing all the part of what makes a personal agent and not having to read openclaw's large codebase.
On demand software generation is going to be as common and foundational in the next 3 years as SaaS is today.
Most actions humans will take online and in some cases in person will cause software to be created.
Going to be wild to see it happen!
Had a call with Founders of Modaic @modaicdev@FaroukAdeleke3 and @ty_todd1
Think of Modaic as huggingface for DSPy. We dive deep into how Modaic solves the data ops problem for AI systems, treating DSPy programs as first-class citizens with proper version control and distribution.
The founders also give us a live demo of the platform, discuss their roadmap for optimising LLM judges ("Evals for Evals"), and share details on their developer tools/packages: the DSPy IntelliSense VS Code extension and DSTS (DSPy for TypeScript).
#dspy #huggingface
Melbourne 🇦🇺 Theme: Forest
Deep sage tones for the Garden City. 🌿 The Hoddle Grid meets the winding Yarra River - this theme makes the concrete jungle look lush and organic.
Singapore 🇸🇬 Theme: Neon Cyberpunk
Future city vibes. 🦾 Electric pinks and cyans turn the island state into a glowing data grid. Perfect for the ultimate smart city.
I saw a poster in a coffee shop and thought, " I could vibe code that." ☕️✨
Introducing maptoposter: A Python tool / script to turn any city into a minimalist piece of art.
First up: New York in the Noir theme. The Manhattan grid that's familiar to all. 🗽
Last Issue of DSPyWeekly for 2025 - Issue 16th
📚 Articles
Stop Writing Prompts Like a Medieval Alchemist: Why it's time to ditch the "alchemy" for programmable modules.
The Meta-Prompting Protocol: Orchestrating LLMs with adversarial feedback loops.
🎥 Videos
DSPy in Rust: Conversation with core contributor Herumb Shandilya.
DSPy.rb: Vicente Reig presents the Ruby-first port at RubyConf.
🚀 Projects
MedSage: A multimodal healthcare assistant with real-time reasoning.
Aisha: A personal shopping assistant powered by Gemini & DSPy.
p.s. Last week of the year historically see less articles because people are busy with Christmas and family as they rightfully should.
See you in 2026.
#DSPy #LLMs #AI #PromptEngineering #SalesforceAI #Rust #Ruby
DSPyWeekly Issue 15 - 12th Dec 2025 is out
This week is absolutely stacked with:
🎙️ Omar Khattab chats with Martin Casado (a16z) on the evolution of Foundation Models @lateinteraction
📖 Early release of Mike Taylor's "Context Engineering with DSPy" book @hammer_mt
🛠️ Building better AI tools with Anthropic's MCP 🧬 @dronathon
Deep dives on GEPA & Compounding Engineering
Plus: An interview on DSPy for Ruby/BAML @highwayvaquero
observability tips, and 7+ new GitHub projects like dspydantic & dspy-toon @mike_pavlukhin
#DSPy #AI #LLMs #AIEngineering
DSPy Signature to break multi-part questions into atomic, self-contained sub-queries, plus classifies them.
Post this you can decide how to handle each query.
P.S. Breaking single query prompt into atomic questions is helpful and needed. The made up broad classification is for demonstration purpose.
#DSPy
6 months ago, we simply did not have folks talking about both @boundaryML (BAML) & @DSPyOSS in the same vein. Now, thanks to the awesome work of @getpy, we're getting content like this.
There's value in understanding both tools! Do watch and share around 🚀
https://t.co/VvQJ3shw1J
DSPy Interview Series kickstarted with @dosco - creator of ax-llm, typescript port of DSPy.
I sit down with Vikram, the creator of AX-LLM, a TypeScript port of DSPy. We talk through his journey in the tech industry, his path into machine learning, and how AX-LLM has evolved over time. Vikram shares some great insights on why signature abstraction matters, how to optimise interactions with models, and the value of community contributions when building software.
We also dig into the contrasts between TypeScript and Python, explore features of AX-LLM ( AxFlow ) and wrap up with his advice for anyone looking to maintain their own language port or open-source project.
#DSPy
Happy Friday Everyone
DSPyWeekly #13 brings you 11 articles, 5 videos, and 4 new projects to explore.
Highlights: ✨ Building Agents with Ephemeral Memory 🛠️ Announcing DSPy Code CLI 📄 New papers on Prompt Optimization
Attention Bengaluru Python Developers
If you want to learn DSPy (An AI application development framework ) or are already a user of it, worth attending the DSPy Meetup.