Workshops: Agentic Coding Tools in Depth
Aider & OpenHands: Workshop on Git-Native Open-Source Agents
Aider and OpenHands stand for two styles of open-source agentic coding: Aider pairs with you in the terminal and commits every change to git, while OpenHands works autonomously in a sandboxed container and can be self-hosted. In this advanced workshop your engineers master both, connect the models of your choice and make agent runs traceable and repeatable.
Who is this workshop for?
- Senior developers and tech leads who live in the terminal and in git
- Platform teams evaluating self-hosted agent infrastructure
- Teams that want automated issue resolution without buying a SaaS agent
- AI engineers who compare models on their own codebase
- Organisations with strict rules on where agent actions may run
What you will take away
- Use Aider for git-native pair programming: targeted file context, repository map, automatic commits and clean undo.
- Run OpenHands self-hosted with a sandboxed runtime and let it work real issues through to a pull request.
- Combine models deliberately, such as a strong model for planning and a faster one for edits, including Claude and open-weight models.
- Wire linters and tests into the agent loop so every change is checked before it reaches review.
- Make agent runs reproducible by pinning versions, configuration and prompts and by recording transcripts.
Agenda
Module 1
Two agent philosophies
- Interactive pair programmer or autonomous agent: which fits when?
- Positioning next to Claude Code, OpenAI Codex CLI and Gemini CLI
- Open-source governance: licences, release rhythm, community
Module 2
Aider in depth
- Adding and dropping files, repository map, conventions file
- Architect and editor modes with different models
- Commit messages, undo, branch strategy and reviewing agent commits
- Exercise: a refactoring with lint and test commands in the loop
Module 3
OpenHands in depth
- Architecture: agent, sandbox runtime, web UI, CLI and headless mode
- Self-hosting: network, resources and access rights
- Exercise: from issue to pull request on a practice repository
Module 4
Model choice and routing
- Connecting Anthropic, OpenAI, Mistral and local endpoints
- Measuring cost and quality on your own tasks
- Fallback models when a provider is unavailable
Module 5
Reproducibility and automation
- Pinned agent and model versions, configuration in the repository
- Scripted runs in CI and batch jobs
- Logging transcripts and diffs for later analysis
Git as the safety net
Both tools build on git, and that is their strength. Aider turns every accepted change into a commit with a descriptive message, so you review, revert or squash agent work like any other history. OpenHands delivers its result as a branch or pull request instead of touching your main line.
We show branch conventions that make agent commits easy to spot, and review habits that stop a long autonomous run from landing as one unreadable diff.
Self-hosting without blind spots
Running OpenHands on your own infrastructure keeps the execution of agent actions in your hands. The model calls still leave the sandbox, though, unless you use local models. Together we sketch the data flows and decide which repositories may use which model endpoint.
- Container sandboxes with restricted outbound network traffic
- No production credentials inside the runtime
- Dedicated service accounts with minimal repository rights
- Resource quotas so runaway tasks stop on their own
Who leads the day
neooutput AI Code Engineers who use these tools in client projects. Hands-on with your repositories or a sandbox copy, remote or on-site, in English or German, with optional coaching while you roll the tools out.
Frequently asked questions
Do we need Docker for OpenHands?
OpenHands runs agent actions in a sandboxed container runtime, so a container environment is the usual setup on laptops and servers. We check your platform requirements beforehand and prepare a suitable environment.
Does Aider work on large codebases?
Yes, if you steer the context deliberately. The repository map gives the model an overview while you add only the files a task really needs. We practise exactly that on a large repository.
How do these tools compare with Claude Code?
Claude Code is a commercial agent from Anthropic, closely integrated with Claude models and features such as hooks and subagents. Aider and OpenHands are open source and model-agnostic: more control over infrastructure, but more setup work on your side.
Workshop
Get in touch
Want open-source agents that fit your git workflow and run on your own infrastructure?
Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.
Prefer to talk first?
15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.
Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)Request a training / workshop
Workshop: Aider & OpenHands: Workshop on Git-Native Open-Source Agents · 1 day
More workshops
- Workshops: Agentic Coding Tools in DepthLocal LLM Coding Agents On-Premise: Workshop for Dev Teams
- Workshops: Agentic Coding Tools in DepthMistral & EU-Sovereign Coding Assistants: Workshop for Dev Teams
- Workshops: Agentic Coding Tools in DepthDelegating Tickets to Async Cloud Coding Agents: Workshop
- Workshops: Agentic Coding Tools in DepthChoosing & Piloting Coding Agents: Workshop for Engineering Leads
- Workshops: Agentic Coding Tools in DepthOne Repo, Many Agents: AGENTS.md, CLAUDE.md & Cursor Rules Workshop
- Deploy, Convert & ScaleScaling a Cursor-Built App: Architecture, Security and Teams
