• Integrate https://github.com/andthattoo/ollama-workflows as a worker. This way, the Admin node can simply pass in a Workflow and the worker will see to it based on that. This provides a really nice generic way of task specification. Resolves feat: use workflows #50
• Make heartbeat response include metadata of provider/model. Users will provide an array of models from env, such as: DKN_MODELS=gpt-3.5-turbo,phi3 from which we can obtain the provider as well.
• Heartbeat response will be at a specific content topic, not /dria/0/uuid/proto but /dria/0/pong/proto.
• Remove synthesis & search_agent workers
• Compute response (i.e. workflow) will be at a specific content topic: /dria/0/results/proto
• Within a workflow task's payload we will have {workflow, model, prompt}.
• start.sh edits to remove --search and --synthesis, and arg for DKN_MODELS (kudos @selimseker)
• Resolves chore: refactor h || s || e logic to just s || e #52
• Resolves chore: remove old Ollama version when it is updated #32 by upgrading Ollama & LangChain
• Closes bug: multiple threads using node.set_busy #34 because there is only one worker (workflows) that will make the node busy
• Resolves Cancellation of Ollama generation #18 by using cancellation during workflow execution
• Based on provider, check if service is online (e.g. Ollama is running / OpenAI key is present)
• Updates Waku to 0.30 and uses RLNv2

Later fixes:

• Resolves bug: too many requests from Compute node to nwaku at first launch during RLN sync #56
• Resolves bug: can run with no models, should give error in that case #59 but instead of giving an error, defaults to OpenAI. Logs this as well.
• Resolves bug: diagnostic peer tracker misses sometimes #57