Mesh: Tools & Docs for AI Models/Workflows

[Proposal setup] Proposal title

Please provide your proposal title

Mesh: Tools & Docs for AI Models/Workflows

[Proposal Summary] Budget Information

Enter the amount of funding you are requesting in ADA

100000

[Proposal Summary] Time

Please specify how many months you expect your project to last

6

[Proposal Summary] Problem Statement

What is the problem you want to solve?

More and more devs using AI models in day to day coding tasks, yet, due to lack of ai-compatible tools & docs, ai models often output low quality making it difficult to use ai for onchain development

[Proposal Summary] Project Dependencies

Does your project have any dependencies on other organizations, technical or otherwise?

No

Describe any dependencies or write 'No dependencies'

No dependencies

[Proposal Summary] Project Open Source

Will your project's outputs be fully open source?

Yes

License and Additional Information

All code will be fully open source on a publicly accessible Mesh github repository

[Theme Selection] Theme

Please choose the most relevant theme and tag related to the outcomes of your proposal.

Developer Tools

[Campaign Category] Category Questions

Describe what makes your idea innovative compared to what has been previously funded (whether by you or others).

The main innovation of this proposal is that for the first time, we do not only address human developers but focus on optimising and updating open source docs & tools for ai-models, capturing the increasing AI capabilities in off and onchain development

Describe what your prototype or MVP will demonstrate, and where it can be accessed.

Our Concept ideally demonstrates how an open source developer project is able to effectively optimise their tools & tooling for ai-models and ultimately drawing from our efforts for their own projects to adapt and adopt ai-tooling to already existing tools like Discord or Code editors & interfaces like VS Code, Cursor etc.

Describe realistic measures of success, ideally with on-chain metrics.

Measurements of success for this project will be ultimately drawn from the use and development activity of our open source code via Github metrics such as number of contributions and contributors, repository dependencies, stars and forks.

Furthermore we will measure success by the number future collaborations we aim to establish during the proposal execution to encourage and support other open source projects on Cardano to use our examples to optimise their own stack for AI-compatibility.

Lastly the usage metrics of the publicly deployed MCP servers demonstrate the direct usage in real time.

[Your Project and Solution] Solution

Please describe your proposed solution and how it addresses the problem

Mimir comprises a set of tools that are geared towards the AI era of web3 coding.

With the Mesh SDK we have seen demand for ready to use libraries that allow building seamless offchain code. Yet, we experince that building applications on Cardano using AI assistance remains challenging due to hallucination and blockchain's inherent complexities.

Hence we are looking at a number of approaches that aim to minimize hallucination as well as enhance LLM's (Large Language Model) abilities to comprehend complexities and their security implications.

Initially this requires improving the documentation on multiple levels. Starting with code interfaces which shall be used by the LLM. Then a broader set of examples that show and explain common use cases in clear standardised formats. Important is that these formats require continued refining to enable meaningful code understanding and utilization by AI systems. This also requires a holistic approach as there is not just one source of information in our decentralized ecosystem. Hence AI optimised public documentation, which is included in the training data for upcoming AI models, lays the foundation.

The next step focuses on Context Engineering and has the goal to optimize the flow of relevant information towards the LLM. A key part of this process is the introduction of Retrieval-Augmented Generation (RAG) as a foundational system, which allows the LLM to pull in accurate, up-to-date documentation at any instance. Instead of relying only on what it learned during the training, the LLM can now retrieve up to date documentation, interface definitions, and examples from a structured knowledge base. This helps to reduce hallucination and ensures that the LLM reflects the current state of Mesh and wider Cardano tooling.

One important improvement in this process is Contextual Retrieval. Unlike regular retrieval methods that only use embeddings or keyword search, Contextual Retrieval adds an extra LLM call during preprocessing. This call generates a short, specific context for each chunk of the document. This context is then added to the chunk before creating embeddings and search indexes. By adding this background information, the system makes sure the LLM gets the most accurate and relevant information. This improves retrieval a lot because it keeps important context that normally gets lost when documents are split into smaller parts.

RAG acts as the link between live documentation and LLM reasoning. Its usefulness depends on how well the underlying material is written and organised, which is why efforts to improve coverage and clarity are so important. As the quality of the documentation improves, so does the LLM's ability to assist accurately. This lays the groundwork for the next stages, where LLMs go beyond suggestions and begin interacting directly with blockchain through MCP servers (Model Context Protocol) and the code they help to write.

Now that the LLM has a solid understanding of coding with Mesh and is able to assist in a helpful manner, MCP Servers come into play. These servers allow an AI agent to call external tools. We want to wrap the existing Mesh functionalities in MCP Server tools. This allows an Agent to interact directly with the blockchain. Initially this is geared towards testing in a testnet environment, but other applications remain to be seen. Additionally MCP Servers act as a bridge into the AI powered IDEs (Integrated development environments), allowing us to serve relevant documentation and additional contextual information during an agent's runtime. Enabling coding agents to make informed and increasingly sensible decisions when building dApps on Cardano.

Finally we do not want to build our solution in a silo! Thus we are focusing on a Collaboration driven approach for building out these tools. In this kick off proposal we will collaborate with No.Witness labs to cross test each other's AI compatibility.

[Your Project and Solution] Impact

Please define the positive impact your project will have on the wider Cardano community

Giving LLMs a rich understanding of the Mesh SDK, will have a significant impact on the developer experience building on Cardano. Mesh is already used by many first time and seasoned Cardano developers and with Mimir these developers will have a powerful AI assistant at their side to help them navigate the numerous complexities.

Furthermore this lays the foundation for deeper AI web3 interaction as AI usage shifts from assisting towards agentic behaviour. With initial steps involving vibe-coding and spec-driven coding agents. Moving towards increasingly autonomous agents directly interacting with blockchains through MCP servers and the code they wrote.

Impact Factors:

Better DevEx

Lower Barrier to entry for new Cardano developers

Groundwork for agentic web3 tooling

Collaboration and exchange of expertise in the ecosystem

Overall envisioned impact is to make it more fast, safe and reliable for AI models to carry out developer tasks, eventually enabling:

Improved onboarding: AI-native devs (the majority of new devs) will get accurate, fast, contextual help when building on Cardano with Mesh.

Increased Developer productivity: Cuts time spent searching docs, guessing UTxO logic, or debugging code while improving code generation and its quality carried out by AI models

Ecosystem-first: Sets a standard that other Cardano SDKs or projects can easily adopt and adapt to improve AI compatibility. Also a great CIP candidate.

Long-term value: Lays the foundation for effective tools for AI agents to interact with, enables new clients, ensures long-term compatibility while accessing new markets

[Your Project and Solution] Capabilities & Feasibility

What is your capability to deliver your project with high levels of trust and accountability? How do you intend to validate if your approach is feasible?

Mesh has been a longtime contributor to the cardano open source developer ecosystem. Our tools are widely used and we have proven our skills and commitment by building and maintaining essential tools which empower many to build on cardano since 2022. Over the time, by non-stop breathing and building Cardano, we have gained some finest expertise which makes it quite likely that we are able to validate and achieve our proposal objectives properly.

We are also proud and glad to have participated in catalyst since Fund 10, and as of today, we are well familiar with the process of completing proposals. For more information about all our funded proposals, their progress and details, we have built a dashboard which makes it easy for you to assess and audit our capabilities of delivering on our promises keeping the style of "don't trust, verify”:

https://gov.meshjs.dev/catalyst-proposals

Furthermore, we already have experience writing documentation and also interactive documentation for human developers. Now we are applying this expertise towards AI driven developers.

This new set of developers has different needs. Hence we are reformatting and enhancing existing material, which is well with our scope of capabilities.

Additionally we have already begun building MCP servers to test our AI integration. This Proposal is meant to scale our efforts towards a comprehensive solution.

[Milestones] Project Milestones

Milestone Title

Optimize Documentation

Milestone Outputs

• Refactored code base featuring ai-optmised inline documentation.

• Improved web docs focusing on scrapability and competability with AI systems

• Create & publish a twitter thread to inform the public

Acceptance Criteria

• Completed Refactoring of code base featuring ai-optmised inline documentation

• Completed web docs page focusing on scrapability and competability with AI systems

• Published twitter thread to inform the public on milestone achievements

Evidence of Completion

• Link to the respective Pull requests adding inline documentation

• Link to the refined web docs page, containing AI ready documentation

• Link to a twitter thread to inform the public

Delivery Month

1

Cost

25000

Progress

20 %

Milestone Title

Build Context Engineering Toolset

Milestone Outputs

• Develop a RAG driven AI assistent embeded in the web docs page.

• Create & publish a twitter thread to inform the public

• Provide URLs to the relevant Pull Requests adding the RAF driven ai-assistent to the code-base

Acceptance Criteria

• Completed web docs integrated AI chat using our RAG solution.

• Published twitter thread to inform the public

• Share link to the relevant Pull Requests adding the RAF driven ai-assistent to the code-base

Evidence of Completion

• Link to additinoal feature "Ask AI" in web docs page.

• Link to twitter thread to inform the public

• Links to relevant Pull Requests adding the RAF driven ai-assistent to the code-base

Delivery Month

2

Cost

25000

Progress

40 %

Milestone Title

Build MCP Servers

Milestone Outputs

• Mesh-providers MCP for querying Blockchain state, utilizing and covering the Blockfrost API.

• Mesh-core MCP for using Mesh's low level features.

• Mesh-transaction MCP for contructing and submitting Transactions

Acceptance Criteria

• Completed Mesh MCP Providers.

• Completed Mesh MCP Core.

• Completed Mesh MCP Transactions

• Published twitter thread to inform the public

• Published links to relevant Pull Requests adding the MCP servers to the code-base

Evidence of Completion

• Link to additinoal feature "Ask AI" in web docs page.

• Link to twitter thread to inform the public

• Links to relevant Pull Requests adding the RAF driven ai-assistent to the code-base

Delivery Month

3

Cost

20000

Progress

60 %

Milestone Title

Refine Flow & Integration Tests

Milestone Outputs

• Comprehensive testing and refining of the software products built in the previous milestones.

• Colaboratory peer testing with No.Witness labs.

• Discord bot to showcase the MCP servers abilities.

Acceptance Criteria

• Completed overview of internal testing efforts.

• Completed Test Report for our partner No.Witness labs.

• Completed Discord bot with small but reasonable rate-limits for bot usage. (AI API usage has significant operational cost and in this only for showcasing puposes.)

• Published twitter thread to inform the public

Evidence of Completion

• Link to Collaboratory Test Reports and Discord bot with on-request test access.

• Link to Twitter thread to inform the public

• Links to relevant Pull Requests adding the milestone outcomes to the code-base

Delivery Month

4

Cost

20000

Progress

80 %

Milestone Title

Wrap up and final Documentation

Milestone Outputs

• Create and submit Proposal Close Out Report

• Create and submit Proposal Close Out Video

• Create Docs and step-bystep-guide for users on how to use the tools

• Create a twitter thread on the milestones completion to inform the public

• License all code on a public github repository under a open source licence

Acceptance Criteria

• Completed the Close Out Report

• Completed the Close Out Video

• Completed how-to-use guides and Docs

• Published twitter thread to inform the public

• Published all code under open source licence

Evidence of Completion

• Link to completed Close Out Report

• Link to completed Close Out Video

• Link to completed guides and Docs

• Link to the project repository on Mesh github with an active open source license

• Link to Twitter thread to inform the public

Delivery Month

6

Cost

10000

Progress

100 %

[Final Pitch] Budget & Costs

Please provide a cost breakdown of the proposed work and resources

The budget breakdown of the proposal deliverables, already outlined in the Proposal Milestones, consists of:

Milestone 1 Budget: 25,000.00 ADA

Milestone 2 Budget: 25,000.00 ADA

Milestone 3 Budget: 20,000.00 ADA

Milestone 4 Budget: 20,000.00 ADA

Milestone 5 Budget: 10,000.00 ADA

[Final Pitch] Value for Money

How does the cost of the project represent value for the Cardano ecosystem?

The clearest way for open-source tools to demonstrate “value for money” is through real usage data by looking at how actively the code is adopted across the developer ecosystem and by which kinds of projects. At Mesh, more than 900 dependent projects currently use our code, with around 10k npm package downloads each month. This positions Mesh as one of the most widely adopted open-source stacks in the Cardano ecosystem, supporting projects across the full spectrum of use cases. These metrics show that the budgets we receive translate directly into solutions that are practical, widely used, and that solve real developer problems.

Funding from this proposal allows us to reward contributors and maintainers of the codebase, while ensuring that all projects depending on Mesh continue to benefit from a rich, constantly evolving feature set—available entirely free, without paywalls, and protected by an open-source license.

For a deeper look into Mesh usage metrics in 2025, see: Mesh Stats.

https://gov.meshjs.dev/mesh-stats

With Mesh Mimir, our goal is to achieve adoption metrics on par with our other open-source tools. As a fully open-source project, Mimir aims to improve the developer experience for projects building on Cardano, making development faster, safer, and AI-ready.

1. A movement, starting with a project

We are in the age of AI, yet using AI for Cardano development remains difficult (tbh, AI currently simply sucks at Web3 coding-tasks). The inherently decentralized nature of the ecosystem means there is no single source of truth (SSoT) for AI to reliably draw from, so, its not rly easy to get AI understanding context of input and expecting high quality output. Mimir is designed to close that gap—not by enforcing a central SSoT, but by providing dedicated tooling that acknowledges AI as part of the community and supports its effective use.

As one of the most widely adopted SDKs in the Cardano ecosystem, Mesh is well-positioned to, indeed, take a leading role in making Cardano accessible to AI models and all developers who are using more and more AI tools to improve their capacities and perfomances. Furthermore, by orienting Mesh towards AI usability, we aim to gain & gather expertise, refine best practices, and share these insights with other projects across the ecosystem while encouraging collaborations to join our efforts to make Cardano development more AI-friendly and effective. Our collaboration with No.Witness Lab is a first spark and we look forward to initiate further collaborations with Cardano dev-projects since no single entity is able to solve the issues we address. Apes strong together, you know.. ^^

[Required Acknowledgements] Consent & Confirmation

Terms and Conditions:

Yes