feat: implement google genai provider #134
Open
+2,183
−1,144
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Test summary |
|
@evalstate I might have botched any semblance of code quality in this PR. Hopefully by moving the Google provider to |
|
THANKYOU!!!! I can't wait to play with this - it looks amazing. I'm going to get the other PRs and a couple of defect fixes released in the next day or 2 and then concentrate on bringing this in. I must confess, I did think implementing this would be quite good fun when you got to the different modalities - how has it been so far? |
Oh don't get your hopes too high; this was an hour and a half work coded with I am pretty confident we've got all of the major features working. The rest will most likely be edge-cases that isn't easily covered by the tests. |
|
Sorry for adding a comment not related to the thread here. But couldn't help ask @monotykamary if you could share the coding agent referred here - "super simple coding agent PoC I'm trying to make" . Quite curios to know how fastagent can be leveraged to make coding agents, that perhaps complements other tools out there including "Cline" that you have mentioned. |
|
OK - I've had a chance to take a quick look at this, and seems to be working pretty well. Having a quick look (and the results of the e2e smoke tests), structured outputs don't seem to be returned: There's also something odd happening with Tool Calls (I've not had a chance to investigate), but if you run the So not sure what's happening with that?! There are a few of the multimodal tests failing, but I suspect this issue above is the root cause. To run the e2e tests (I've added Below is also some code if you just want an agent that uses structured outputs: Anyway, thank you for the contrib so far - just ping me on here/discord if you need any help/assistance with this :) |
|
Oh I'll have a quick look 🚀 |
monotykamary
force-pushed
the
feat/implement-google-genai-provider
branch
from
fca0262 to
4aa8dd1
Compare
@rkunnamp An exercise up to the reader... I'm kidding, it's really abusing the DesktopCommander MCP and formatting the agents in a way that pass history to each other. One example would be using an infinite loop and pass history whenever import asyncio
from mcp_agent import RequestParams
from mcp_agent.core.fastagent import FastAgent
...
async def main():
async with fast.run() as agent:
while True:
# Start an interactive session with architect
await agent.interactive(agent="architect")
# Pass architect's history to coder for generation
await agent.coder.generate(agent.architect.message_history)
# Start an interactive session with coder
await agent.interactive(agent="coder")
# Pass architect's history to coder for generation
await agent.architect.generate(agent.coder.message_history)
if __name__ == "__main__":
asyncio.run(main())There are much better patterns that you can use to inject it directly into a following user prompt, instead of updating the system prompt like this to avoid:
Really, the key player here is the MCP server and the agent architecture is up to you. skydeckai-code, or originally called |
evalstate
reviewed
May 4, 2025
| systemPrompt=self.instruction, | ||
| parallel_tool_calls=False, | ||
| systemPrompt=self.instruction, # System instruction will be mapped in _google_completion | ||
| parallel_tool_calls=True, # Assume parallel tool calls are supported by default with native API | ||
| max_iterations=10, |
|
That's great - looks like structured generation is really solid now. And these models are fast through this native interface - really cool :) I've done another pass of testing and:
Anyway, great stuff so far - think this is a huge enhancement to fast-agent - thanks! |
…tracking for Gemini models
monotykamary
force-pushed
the
feat/implement-google-genai-provider
branch
from
a9c2b21 to
6907b1c
Compare
…le-genai-provider
….com/monotykamary/fast-agent into pr/monotykamary/134-1
….com/monotykamary/fast-agent into pr/monotykamary/134
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR implements the Google provider for fast-agent using the native
google.genailibrary and should resolve #6 .Key changes include:
google.genai.Clientfor interacting with Google's generative models.PromptMessageMultipartandgoogle.genai.types.Content.GoogleConverterclass to handle data structure conversions.Image test code sample
This work is a step towards fully leveraging the features of the
google.genailibrary within fast-agent.