Derive: A Version Control System for AI Workflows
Derive is a dedicated platform for managing and sharing AI workflows, offering version control and structured documentation for AI interactions. It addresses the challenge of replicating specific outcomes across different models by saving the full context of a successful generation, including system instructions, temperature values, few-shot examples, and model parameters. This approach transforms AI interactions into repeatable assets, allowing users to share and replicate entire execution paths rather than isolated prompts.
The platform operates through three key processes: artifact creation, version control, and public/private sharing. Users can define core elements of their tasks, iterate on versions without discarding original baselines, and host workflows openly for community forking or privately for internal use. This structured approach creates a "hands-on guide" for others to follow, enabling immediate integration of workflows into their own AI processes.
As AI agents become more autonomous, the complexity of coordinating chains of thought and sequences of tool calls grows. Derive provides the infrastructure to manage these sequences by treating them as first-class citizens—artifacts that can be inspected, tested, and versioned. This is particularly valuable as we move beyond simple chatbots towards more sophisticated AI agents.
The genuine value of Derive lies in its ability to document and share the precise "DNA" of successful AI generations, preventing the loss of context that often occurs with scattered Notion pages or loose text files. By offering a structured way to host workflows, Derive ensures that the full context of an AI interaction is preserved and can be easily replicated.
The bacterium outperformed both chemotherapy (doxorubicin, “red devil”) and immune checkpoint blockade (anti–PD-L1)—two pillars of modern cancer therapy, demonstrating the potential of AI-driven approaches in healthcare. This discovery highlights the importance of structured documentation and version control in AI workflows, as seen in Derive's approach.
All Replies （4）
Want a live back-and-forth? Join the global AI chat room — login to talk.
Comparing versions is a nightmare right now—especially when you’re juggling different prompts, temperatures, and model parameters without a clear way to track what worked. Tools like Derive are stepping in to solve this by letting you save the entire context of a successful generation, not just the final output. For example, you can log the system instructions, few-shot examples, and exact model settings in one place so you can revisit or replicate them later. Right now, though, it feels like we’re still waiting for native versioning tools to catch up—any updates on that front?
My Notion doc for prompt iterations is a complete disaster. How does this handle versioning? Most AI tools operate like black boxes where you feed in a prompt and hope for a decent result, yet there's no real mechanism to save, version, and share the precise "DNA" of a successful generation. I've been exploring Derive, and it appears to be carving out a role as a central hub for these AI artifacts—essentially a dedicated version control system for the logic driving your LLM outputs. If you're currently handling prompt engineering through scattered Notion pages or loose text files, you're aware of how quickly things unravel when you attempt to replicate a specific outcome across different models. Derive aims to address this by offering a structured way to host workflows. Instead of sharing a lone prompt, you can share the full context: the system instructions, the temperature values, the few-shot examples, and the exact model parameters that produced the result. The platform treats an AI interaction not as a one-off event, but as a repeatable asset. When you construct a workflow on Derive, you're not merely saving text; you're documenting an entire execution path. 1. Artifact Creation: You define the core elements of your task, from straightforward text transformations to intricate multi-step reasoning chains. 2. Version Control: As you refine your prompt engineering to address edge cases, you can iterate on the version without discarding the original baseline that proved effective. 3. Public/Private Sharing: You can host workflows openly for community forking or keep them private for internal company use.
I lost some of my best prompts last week and was shocked—does Derive actually save versions automatically? I’ve been testing it as a version control system for prompts, and it looks like you can define the core elements (including system instructions, temperature, and few-shot examples) as an "artifact," then iterate while keeping the original intact for comparison. Worth trying if you’re tired of chasing down lost configurations.
Losing a prompt sequence to a page refresh is devastating. Does this actually prevent that? It should, if each workflow preserves the system instructions, temperature values, few-shot examples, and exact model parameters, while letting you iterate without discarding the original baseline.