You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Description
We want to expand Hive’s agent definition so that each agent can declare which framework it uses (Bee Framework or smolagents) in a single YAML config. This would allow developers to choose their preferred agent framework without maintaining separate definitions or code paths.
Proposed Approach
Agent Type Field
In the YAML file, add a field like type: "bee" or type: "smolagent".
If type: "bee", instantiate a BeeAgent from the Bee Framework.
If type: "smolagent", instantiate a smolagent-based agent.
Single LLM Per Agent
Each agent remains tied to just one LLM (e.g., “IBM Granite,” “Llama 3.x,” etc.). If multiple LLMs/backends are needed, we deploy multiple agents.
Unified Instantiation Logic
Hive’s internal code checks the YAML and branches accordingly.
The rest of the agent’s config (name, instructions, tools, etc.) stays consistent regardless of the type.
Future Extensions
Later on, we can add advanced collaboration or multi-LLM capabilities. For now, the focus is on a clean and simple approach to define either Bee or smolagents from one file.
Acceptance Criteria
A single YAML definition where the user can specify type: "bee" or type: "smolagent" for each agent.
A provisioning or creation script in Hive that reads the agent type and instantiates the correct agent class.
Basic validation/error handling if type is missing or unrecognized.
Reference
Relates to Issue agent: definition #47 (Agent Definition), where we discussed supporting multiple backends (one per agent).
If this approach looks good, I’ll start drafting a PR and share it for review!
The text was updated successfully, but these errors were encountered:
Description
We want to expand Hive’s agent definition so that each agent can declare which framework it uses (Bee Framework or smolagents) in a single YAML config. This would allow developers to choose their preferred agent framework without maintaining separate definitions or code paths.
Proposed Approach
Agent Type Field
type: "bee"
ortype: "smolagent"
.type: "bee"
, instantiate a BeeAgent from the Bee Framework.type: "smolagent"
, instantiate a smolagent-based agent.Single LLM Per Agent
Unified Instantiation Logic
type
.Future Extensions
Acceptance Criteria
type: "bee"
ortype: "smolagent"
for each agent.type
is missing or unrecognized.Reference
If this approach looks good, I’ll start drafting a PR and share it for review!
The text was updated successfully, but these errors were encountered: