Connect bluebox agent and browser agent #160
Conversation
rayruizhiliao
changed the title
rayruizhiliao
changed the base branch from
main
to
agent-to-use-semantic-search
dimavrem22
force-pushed
the
agent-to-use-semantic-search
branch
from
cfefe46 to
ea6aa46
Compare
bluebox/agents/bluebox_agent.py
Outdated
| timeout_seconds: int = 300, | ||
| use_vision: bool = True, |
There was a problem hiding this comment.
the agent prob doesnt need to know about these. lets remove them from params and hardcode in the endpoint
bluebox/agents/bluebox_agent.py
Outdated
|
|
||
| try: | ||
| with requests.post( | ||
| f"{Config.VECTORLY_API_BASE}/buagent/execute/stream", |
There was a problem hiding this comment.
lets rename this endpoint to browser_agent instead of buagent
| continue | ||
|
|
||
| try: | ||
| data = json.loads(line[6:]) |
There was a problem hiding this comment.
can we have data models that mirror those in servers for this? so that we can use their fields and not .get(...) when parsing them below
| Progress is streamed in real time via SSE. | ||
|
|
||
| Args: | ||
| task: Detailed natural language instruction for the browser agent. Be specific and step-by-step. |
There was a problem hiding this comment.
lets add an example instruction to convey the level of detail we need. either here or in system prompt
There was a problem hiding this comment.
can we do it in the upcoming the bluebox "finalization" PR? because i don't yet have the big picture of the bluebox agent, and how it would pass information from the initial user conversations -> routine search -> converted task description? Also i think we need to play with the browser agent for a bit to give good examples. Added a TODO in commit 3ef2ed5
2 participants
New tool:
_execute_browser_taskPOST {VECTORLY_API_BASE}/buagent/executewith the sameX-Service-Tokenauth pattern used by the existing routine toolstask(natural language), timeout_seconds (default 300, clamped to 30-1800), anduse_vision(default True)timeout_seconds + 30so the HTTP call doesn't expire before the agent doessuccess,final_result,errors,n_steps, etc.Updated system prompt
execute_browser_taskTODO