A funding opportunity for organisations across sub-Saharan Africa that have completed GAIC training and want to adapt an existing tool for their communities.
Imagine a women-led market traders’ association adapting an existing tool to help members find out how to get a market stall. Members could ask a question by voice in the language they use at the market and find out what documents they need, what the official fees are and where to apply. The traders would decide which questions the tool should answer and test whether the answers help them take the next step.
The A+ Alliance and the Gender and AI Innovation Collective (GAIC) are inviting women’s rights, feminist and other civil society organisations that have completed GAIC training to put that learning into practice. Through this funding call, teams can adapt an existing open-source tool, resource or AI solution. That could mean adapting a tool you already know, building on a prototype from the first GAIC cohort, or using AI to make an open-source resource easier to use.
What you can do in the Sandbox
Created by the A+ Alliance, the GAIC AI Sandbox, run by Code for Africa, gives grantee teams shared technical infrastructure for building and testing their projects. Teams can try different AI models, work with their own reference materials in a team-specific document library and manage model use within an allocated budget. They can also draw on the shared feminist, decolonial collection curated by Women at the Table and connect with other teams through the Circle Community.
In the Sandbox, the association could compare how different AI models answer questions using market rules, fee schedules and application forms it has checked with the local authority. Members could ask the same question in different languages or switch between languages as they normally do. The team could then check whether the tool gives the correct fee, explains the next step clearly and directs people to the right office when the guidance is unclear.
The Sandbox gives teams a place to do that work. The organisation and its community decide what the tool should do, which sources it should draw on and what it must get right before people rely on it.
Before you choose a tool:
• Start with the people who would use it. What do they want it to help them do? Who could face barriers or be harmed by a wrong answer? How will they help shape and test it?
• Check what you can reuse. Does the licence allow the changes and redistribution you have in mind? Can you access the code, model, documentation and data you need?
• Work out what needs to change locally. Which languages, laws, services and cultural contexts matter? Who can review the local content and keep it current?
• Test what matters before choosing. Use realistic tasks to check language quality, accessibility, bias and safety.
• Plan for data, costs and maintenance. What information do you need to collect or store, who can access it, and who will maintain the tool and cover running costs?
Several of these points depend less on the tool itself and more on the infrastructure it runs on: where data is stored, who pays for computing, and how easily you can test different models. This is where the GAIC AI Sandbox comes in.
Where GAIC teams build: inside the AI Sandbox
Every team in the Gender & AI Innovation Collective builds its own product, such as a conversational agent, in its own language and for its own community. Behind each product sits the same shared foundation: the AI Sandbox. It gives every grantee team the same set of AI tools, its own private document library and a spending allowance. Organisations can focus on their communities, with the technical infrastructure already in place. Think of it as one shared building.
How the Sandbox connects to the tools
• The same model families. The Sandbox offers 45 tested open source models hosted on Amazon Bedrock, from companies including Meta, Mistral, Google, OpenAI, DeepSeek and Qwen. These include the Llama and Qwen families that UlizaLlama and Sunflower are built on, so teams can see how these families handle their own language and task. Switching models takes a one-word change.
• Libraries for local knowledge. Each cohort shares a reading room of documents curated by Women At The Table, and each team has a private library for its own material. This is where the national legal guidance, reporting pathways and information tools locally useful can live. The AI looks up relevant passages before it answers, a method called retrieval-augmented generation (RAG).
• A predictable budget. Each team’s allowance refills on a schedule, so no organisation can run up a surprise bill.
• A shared community. The circle community shares information on teams building something similar, so a GBV chatbot team in Nairobi can find a legal-rights team in Lagos.
Frequently asked questions
What does “gender-responsive AI” mean?
Gender-responsive AI takes into account how gender intersects with race, class, language, disability and age to shape what people need from a tool, whether they can use it and what risks they face.
Can a small CSO localise an AI tool without in-house engineers?
Yes. Through GAIC, the AI Sandbox provides the infrastructure, models and document libraries, so teams can focus on content, prompts and community testing.
How do I apply to localise a tool?
Read the full GAIC open call and gather your organisation details, the link to the open source tool and a short proposal before you start.
From training to a proposal
If your organisation has completed GAIC training, choose one problem your community wants to solve. In your proposal, tell us what you would adapt, what would need to change, who would help shape it and how you would know it works for the people who will use it.
See the GAIC funding call and Submit a proposal
Cover Photo: Luke Conroy and Anne Fehres & AI4Media / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/