.png)
Cedar helps climate-related companies build AI automations that streamline data work and sustainability reporting. They help clients build workflows that automate carbon accounting, carbon project risk assessments, and due diligence and RFP questionnaires, delivering real AI productivity gains beyond chatbots.
Unlike traditional sustainability software, which relies on rigid, predefined workflows, Cedar provides a set of baseline AI workflows that can be adapted to each company’s needs. These workflows simplify human verification by producing auditable outputs with human sign-off for critical decisions. The result is AI-human collaboration that truly automates a company’s proprietary processes, helping businesses gain a competitive edge with AI.
They offer clients something unique: a flexible, end-to-end system that replaces manual effort with intelligent execution, not just tooling.
However, these AI-powered capabilities are only as effective as the agent framework behind them.
Early on, Cedar built its solution using LangChain, which made it possible to prototype agent behavior and validate early use cases. But as Cedar’s platform matured and customer needs became more complex, the limitations of a general-purpose framework became clear. The Cedar team realized that they needed to rethink the framework underneath their AI system.
We spoke with Ravish Rawal, Head of A.I. Engineering at Cedar, about how their platform evolved, why they moved from LangChain to Agno, and what it takes to run a multi-agent system in production.
Why did Cedar move from LangChain to Agno?
Building on LangChain, the Cedar team began to notice multiple issues:
- different models expected different message formats
- documentation was sparse and outdated
- flexibility was limited, making it hard to adapt and scale
- speed suffered even with small changes
As a result, the Cedar codebase became increasingly complex and difficult to maintain, resulting in growing technical debt. These limitations prompted the team to look for a different solution.
Agno addressed these challenges directly:
- Agno lets teams abstract away model-specific quirks while keeping workflows consistent, maintainable, and easily swappable between models.
- Agno provides clear, up-to-date, and searchable documentation to help developers move quickly.
- Agno’s modular design allows you to swap LLMs, databases, or vector stores without rewriting code.
- Agno is the fastest agent framework on the market.
Cedar made the switch in April 2025, and they “never looked back.”
According to Ravish, migrating the Cedar codebase to Agno was smooth and efficient, allowing the team to become production-ready in about a week while immediately benefiting from faster iteration, greater flexibility, and reduced technical debt.
How did Agno supercharge Cedar’s platform?
According to Ravish Rawal, the move from LangChain to Agno allowed Cedar to stand out amid the surge of AI-powered sustainability tools:
Smarter solutions
“Not only could agents carry history into context, but we could also customize how much history, and even have this be agentically determined, without building these tools ourselves. All of these were controlled by booleans, which made it super easy.”
Greater flexibility and control
“RAG was also very easy & flexible with Agno. We were able to activate hybrid search, vector search and reranking all with boolean switches. Easy to develop and to benchmark”
Faster debugging and iteration
“We loved how transparent and easy it was to debug using AgentOS, which allowed us to ship much quicker.”
What was the hardest production challenge they faced?
Together with Agno, Cedar has tackled multiple sophisticated production challenges, including context window management, hybrid search with document filtering, parallel agent coordination, and session state management. But what's been the hardest technical problem they’ve solved?
According to Ravish, it was a challenge brought about by rapid growth. “We found ourselves needing to not only process hundreds of user documents, but also simultaneously run computation on them and combine it all with 3rd party data.”
To stay ahead, they evolved from dumb-retrieval to a platform that can process large folders and databases in one shot, performing complex analyses, classifications, and computations. To make that work, Cedar uses the full spectrum of Agno’s offerings: a layered architecture of teams, agents, tools, and workflows. They’ve also optimized their retrieval using a custom algorithm that Agno agents can accommodate out of the box.
What advice would they give to engineering teams?
“Gather your eval sets as you go.”
Many modern evaluation methods, like black-box evaluations or LLM-as-a-judge, only work well if the test inputs reflect real user behavior rather than idealized or made-up examples. Real users create messy inputs, edge cases, and stress scenarios that are difficult to predict in advance and are specific to your product.
If you try to sit down later and invent an eval dataset from scratch, it is much harder and less accurate. If you instead capture examples as they naturally occur through user feedback, failed runs, odd outputs, support tickets, or internal testing, you end up with evals that better reflect production reality.
Building together: How Agno and Cedar continue in partnership
Agno and Cedar share a focus on flexible solutions and practical outcomes. From the start, we’ve collaborated closely to lay the groundwork for the next phase of innovation.
Many of the capabilities Cedar relies on today were added in response to conversations with the Agno team.
For example, Cedar’s team noticed that different AI models excel at different tasks: some are better at tool-calling (invoking APIs, running calculations, or retrieving external data), while others are better at interacting with users (synthesizing information, giving instructions, or suggesting follow-up actions). They requested the ability to use different models in tandem rather than relying on a single model for everything, for example using specific LLMs for tool-calling and others for synthesis, so they wouldn’t have to compromise one strength for another.
The Agno team shipped output_model in 24 hours. By separating backend intelligence from frontend interaction, it lets Cedar use the best model for each task while keeping the experience seamless for users.
Key benefits include:
- Improved recall in RAG workflows: the system retrieves the right information more reliably.
- Greater depth across all functions: tasks are executed more intelligently because the most capable models are used for each step.
- Faster responses: tools are called correctly on the first attempt, reducing retries.
- Consistent user experience: users continue to interact with Cedar AI in the way they expect, even as underlying models change.
We are excited to continue this close and responsive partnership with Cedar, an industry leader in climate-related AI automations. By staying aligned on innovation and customer needs, we aim to keep delivering flexible, practical solutions that push the boundaries of what their platform can achieve.
What can you build with Agno?.