Community Roundup: December 2025

Yuvaraj Shanmugam
December 30, 2025
8 min read

December 2025 continued Agno’s rapid growth, with 19 major releases, 120+ contributors, and major features like RBAC, native tracing, and HITL improvements.

Whether you’re deploying multi-agent systems to production, managing user permissions across teams, or building conversational workflows with human oversight, December delivered the tools you need to ship with confidence.

Plus, we welcomed 180+ new community members to our Discord.

New releases & features

v2.3.0 - v2.3.19

December delivered nineteen major releases. Here's what we shipped:

AgentOS Role-Based Access Control (RBAC) (v2.3.13)

The flagship feature of December: enterprise-grade authorization for AgentOS. RBAC enables:

  • JWT-based authorization — All traffic requires signed tokens with correct permission scopes
  • Per-endpoint authorization — Grant or deny access based on JWT scopes
  • Per-agent resource control — Control which users can run or view specific agents using scopes like agents:my-agent:read
  • Team and workflow control — Same granular access for teams and workflows

This transforms AgentOS from a development tool into a production platform that lets you safely expose AI agents to end users with fine-grained permissions.

Native tracing in Agno (v2.3.5)

OpenTelemetry-based observability is built directly into Agno. Automatically captures:

  • Agent and team execution details
  • Model calls and token usage
  • Tool executions and results
  • Team operations and delegation

All stored in your Agno database of choice, with no external services required. Perfect for debugging production issues and understanding agent behavior.

HITL (human-in-the-loop) improvements (v2.3.7)

Completely revamped developer experience for human oversight:

  • New RunRequirement class for handling HITL flows
  • Available in Agent response and RunPaused event when streaming
  • Resolve requirements directly via requirement objects
  • Better support for workflows containing agents with HITL

Essential for production applications that require human approval before critical actions.

Reasoning model streaming (v2.3.14)

Stream reasoning chunks in real-time when using a dedicated reasoning_model. Users can now see the agent's thinking process as it happens, which is critical for transparency and debugging complex reasoning flows.

Remote agents (v2.3.17)

New RemoteAgent, RemoteTeam, and RemoteWorkflow classes enable the usage of agents running on a remote AgentOS. Plus, the AgentOSClient class for programmatic access to remote AgentOS instances. Build distributed agent systems that span multiple deployments.

Context compression (v2.3.3)

Manage agent context while running by compressing tool call results. Helps agents stay within context windows and avoid rate limits. Includes token counting support (v2.3.12) for precise compression control. Still in beta but production-ready.

Memory optimization (v2.3.3)

Summarize user memories outside of agent runs using optimize_memories and aoptimize_memories on the MemoryManager. Keeps memory stores lean and relevant as conversations grow.

Database migrations manager (v2.3.0)

Introducing the MigrationManager class, together with our first migrations for sessions and memories tables. Essential for teams using PostgreSQL, SQLite, MySQL, or SingleStore who need to keep schemas up-to-date.

Async MySQL support (v2.3.9)

Added AsyncMySQLDb with compatibility for the asyncmy driver. Full async support now spans PostgreSQL, MySQL, MongoDB, and SQLite.

AgentAsJudge evaluations (v2.3.9)

LLM-as-a-judge evaluation system with:

  • Binary (pass/fail) or numeric (1-10) scoring
  • Standalone runs, post-hooks, or background execution
  • Custom evaluator agents
  • Support for custom criteria

Perfect for automated quality assurance in production.

Gemini file search support (v2.3.3)

Native integration with Gemini's file search capabilities:

  • Document store management (create, list, get, delete)
  • File upload with custom chunking configuration
  • Citation extraction from responses
  • Full async support

Check out the cookbooks: cookbook/models/google/gemini/file_search_*.py

A2A interface enhancements (v2.3.14)

Updated A2A interface with new endpoints:

  • Get an Agent Card for any agent, team, or workflow
  • Updated run endpoints matching the latest protocol
Hooks as background tasks (v2.3.5)

Pre- and post-hooks can now run as background tasks on AgentOS, enabling non-blocking execution. Perfect for expensive operations like notifications, logging, or evaluations that aren't on the critical path.

ChromaDB hybrid search (v2.3.17)

Combine dense vector similarity search (semantic) with full-text search (keyword/lexical) using RRF fusion for local ChromaDB deployments. Best of both worlds for knowledge retrieval.

New toolkits
Model updates
JSON schema for structured outputs (v2.3.14)

output_schema now accepts JSON schemas in provider-specific formats, passed directly to model APIs without transformation. Full control over structured output configuration for OpenAI, Claude, and compatible providers.

Community projects & showcases

Our community built incredible things in December. Here are the standout projects:

1. Multi-project showcase by Alexandre E. Souza

Alexandre continued his prolific contributions with multiple projects spanning education, entertainment, and developer resources:

agno-golang port & book

English teaching agent

A local-first agent that runs entirely offline using Agno + Ollama. Perfect example of privacy-first AI education tools. LinkedIn post

Emma the VTuber

Created Emma, an AI-powered VTuber using Agno for conversational intelligence. Demonstrates Agno's versatility beyond traditional agent applications. LinkedIn post

2. Healthcare & entertainment applications by Touseef Shaik

Prescription parser & dosage manager

Upload prescriptions and the app:

  • Extracts medicine names accurately
  • Figures out dosage schedules (morning, afternoon, night)
  • Translates summaries into your local language

Perfect example of Agno in healthcare accessibility. LinkedIn post

Cast relationship graph

Built a visualization tool for understanding character relationships in media. LinkedIn post

3. Educational content series by TechieTalksAI

Web-powered AI research agent

Comprehensive tutorial showing how to build research agents with web capabilities. YouTube

Gemini + Streamlit + uv tutorial

Beginner-friendly guide to building AI agents with modern tooling. YouTube

4. Personal finance copilot by Build Fast with AI

AI-powered financial planning assistant built with Agno. Complete with a deployed app and open-source code:

5. Universal support bot by Raghavender Reddy Grudhanti

Enterprise support system using Agno + OpenAI + Streamlit that automatically routes to Salesforce, Jira, and ServiceNow. A perfect example of using Agno to unify multiple enterprise systems behind a single conversational interface. Read the full article.

6. vLACQ stack by Kameshwara Pavan Kumar Mantha

Production-ready RAG architecture combining vLLM + LlamaIndex + Agno + Chonkie + Qdrant. Also published a detailed RAG implementation with visual intelligence:

7. SlideshotsHQ auto-wizard by Tanpreet

Introducing the auto-wizard feature for SlideshotsHQ, powered by Agno's agentic capabilities.

8. AI investment advisory agent by Sowunmi Oluwamayowa

Financial advisory agent helping users make informed investment decisions. Twitter showcase

9. Production-ready AgentOS tutorial by Eddy Giusepe Chirinos Isidro

Step-by-step guide for creating production-ready AI agents with AgentOS, covering deployment patterns and best practices. LinkedIn post

10. Memory-powered AI agent tutorial by Ihtisham Mehmood

Video guide on building agents with persistent memory using mem0 + Agno. YouTube

11. "The FASTEST agentic framework explained in 14 minutes" by Tyler AI

14-minute deep-dive into why Agno is built for speed and production use. YouTube

12. AgnoGuard by aayush598

Comprehensive LLM guardrails library for the Agno framework, with 50+ safety measures for production AI applications. Open-source and ready to use. GitHub

Contributor shoutouts

December saw 120+ contributors across 19 major releases. Special recognition to our community contributors:

Top code contributors
  • @anurag12-webster (GitHub) — SFX support for ModelsLab
  • @Arindam200 (GitHub) — Nebius rebranding
  • @fkwfung (GitHub) — Gemini 3.0 thought signatures
  • @AkhilmsAchu (GitHub) — YFinance session parameter
  • @Forget-C (GitHub) — RedisCluster support
  • @Sebastiao-Assuncao (GitHub) — Beta types streaming handling
  • @PhysicallyActive (GitHub) — Register MCP subclasses
  • @bhatt-neel-dev (GitHub) — NanaBanana toolkit
  • @Bessouat40 (GitHub) — Broken docs links fix
  • @naaa760 (GitHub) — Bedrock API key auth
  • @esw-amzn (GitHub) — AWS Claude model improvements
  • @aviscode (GitHub) — Team JSON output prompt logic
Feature implementers
  • @suguruxdub (GitHub) — RunEvent serialization fix
  • @thegeosman (GitHub) — Dynamo schemas update
  • @philschmid (GitHub) — Google Gemini 3 README revision
  • @saarya-cyera (GitHub) — ToolExecution timestamp fix
  • @AlanPonnachan (GitHub) — Team metrics duration and timer
  • @opleonnn (GitHub) — Empty dict references preservation
  • @xpander-moriel (GitHub) — AsyncPostgres table creation
  • @RogerHYang (GitHub) — Context variable propagation
  • @Boburmirzo (GitHub) — Memori SDK update
  • @maorkipnis (GitHub) — Nested steps in Parallel groups
  • @edlsh (GitHub) — HackerNews unit tests, Bedrock credential validation, RunRequirement serialization
  • @pengye91 (GitHub) — DeepSeek thinking mode with tools
  • @zhiheng-yu (GitHub) — OpenRouter reasoning support
  • @alikhtag (GitHub) — run_id parameter
  • @VirusDumb (GitHub) — A2A interface endpoints
  • @kyleaton (GitHub) — Documentation link updates
  • @catarinaopires (GitHub) — OpenRouter cost metrics
  • @Marincor (GitHub) — Gemini credentials args
First-time contributor

In the last few weeks, we've extended a warm welcome to 30+ first-time contributors, including @anurag12-webster (SFX support for ModelsLab), @Arindam200 (Nebius rebranding), @fkwfung (Gemini 3.0 thought signatures), @AkhilmsAchu (YFinance session parameter), @PhysicallyActive (MCP subclasses registration), @naaa760 (Bedrock API key auth), @esw-amzn (AWS Claude improvements), @suguruxdub (RunEvent serialization), @thegeosman (Dynamo schemas), @saarya-cyera (ToolExecution timestamps), @AlanPonnachan (team metrics), @opleonnn (dict references), @xpander-moriel (AsyncPostgres table creation), @RogerHYang (context variable propagation), @maorkipnis (nested Parallel steps), @edlsh (HackerNews tests and Bedrock validation), @pengye91 (DeepSeek thinking mode), @zhiheng-yu (OpenRouter reasoning), @alikhtag (run_id parameter), @catarinaopires (OpenRouter cost metrics), and many more. Your PRs, bug reports, and feature requests are the lifeblood of this project.

Community MVPs
  • @aberk — Continued exceptional contributions with detailed bug reports, async database debugging, structured output testing, and production feedback across multiple releases
  • @HudyWeas — 40-message thread debugging tracing setup, helping establish best practices for observability
  • @gigaliths — Critical eval connection error debugging that helped identify async handling issues
  • @supratikb — Deep questions about HITL in workflows, pushing the boundaries of conversational workflow patterns
  • @Shlomi — Persistent advocacy for team delegation improvements and PGVector custom schema support
  • @Griflet — Continued advocacy for session persistence with hook failures—critical for production
  • @yns — Working with the community to understand fine-tuned model integration patterns

What's next

On the roadmap
  • HITL for workflows - Introducing HITL support for all main workflows scenarios
  • HITL for teams - Introducing HITL support for all main Team scenarios
  • Extended vector database support - More databases based on community demand
Get involved

We're always looking for contributors! Areas where you can help:

  • Documentation — Help new users get started faster
  • Integrations — Build connectors for your favorite tools and services
  • Cookbooks — Share real-world patterns and use cases
  • Bug reports — Production feedback makes Agno better for everyone

Check out open issues or join the Discord to connect with the team.

Join the movement

December 2025 proved that Agno is ready for enterprise production. From RBAC to native tracing, from HITL improvements to remote agents, every feature shipped this month answered a critical production need.

Whether you're prototyping your first agent or deploying multi-agent systems at scale, Agno gives you the tools to move fast without breaking things. And with a community this vibrant and growing this quickly, you're never building alone.

Keep building, keep shipping, keep pushing the boundaries of what agents can do.

— Team Agno

‍