Replies: 4 comments 10 replies
-
|
goose is designed as a general purpose engineering assistant that can handle both code and non-code tasks, while Claude Code is more specialized in code-specific operations. goose is highly extensible, allowing you to use any LLM of your choice, not just Anthropic's. In fact, you can even use Claude Code as an LLM Provider inside goose! so basically goose gives you everything Claude Code does plus local control, more flexibility, broader integrations, and full automation.
|
Beta Was this translation helpful? Give feedback.
-
|
that is exactly it @endolith |
Beta Was this translation helpful? Give feedback.
-
|
use any LLM-proxy and CC will work with any model, like gpt-5-codex |
Beta Was this translation helpful? Give feedback.
-
|
this analysis is outdated |
Beta Was this translation helpful? Give feedback.
-
|
From a architecture of goose is there any ability in terms of less token consumption or other approaches that makes its better to use compared to some IDE agent solutions for code , i am trying to understand what are the technology pros that makes Goose stand out |
Beta Was this translation helpful? Give feedback.
-
|
@brijoobopanna it is such a moving target it is so hard to be specifc - IDE agent solutions can be similar as well (ie goose can run via ACP in IDEs), likewise cursor has a cli. I guess it is more app difference (goose has a desktop app making a bit closer to claude desktop or chatgpts desktop, but with more extensions) - and designed to be adaptable. I don't think that means anything for token consumption though. Curious what you are looking for information wise - ie are you looking to choose or try things, make a decision or recommendation or just curious to explore? |
Beta Was this translation helpful? Give feedback.
-
|
One more info please : |
Beta Was this translation helpful? Give feedback.
-
|
only some old ones. https://www.tbench.ai/ is what we use at the moment - often we can store up near top there, but working on getting a regular weekly cadence of it to track performance (as much as it is worth, but that test seems fairly helpful and practical). @tlongwell-block has been looking at that (along with me) - using models like opus 4.1, or sonnet 4.5 (and even some ones like GLM 4.6 which are open source - but usually 10% behind the leaders) |
Beta Was this translation helpful? Give feedback.
-
|
Thankyou I am looking more of a function and performance comparison as the
world of agentic development has many options such as codex, Claude code,
warp and goose, from a Open source perspective goose is promising how about
on performance security and other params
…On Mon, 17 Nov 2025 at 4:27 AM, Michael Neale ***@***.***> wrote:
@brijoobopanna <https://github.com/brijoobopanna> it is such a moving
target it is so hard to be specifc - IDE agent solutions can be similar as
well (ie goose can run via ACP in IDEs), likewise cursor has a cli. I guess
it is more app difference (goose has a desktop app making a bit closer to
claude desktop or chatgpts desktop, but with more extensions) - and
designed to be adaptable. I don't think that means anything for token
consumption though. Curious what you are looking for information wise - ie
are you looking to choose or try things, make a decision or recommendation
or just curious to explore?
—
Reply to this email directly, view it on GitHub
<#3133 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AETH734EBRTSADALUZMPAOT35D6OHAVCNFSM6AAAAACAKDHENCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIOJYGM4DINA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
we are trying to regularly run the "tbench" benchmark, which shows things as competitive (it is not always easy to run it continuously), especially with similar models (we also want to ensure it performs well with open models). So performance I think can be similar (I think you will find many things converge there, so then becomes preference based). On security - some take an approach of asking permissions, fine grained etc, goose has a few modes of operation from chat only, permission but my favourite (and I expect most common) is the autonomous mode - which does make you think differently about risk and security. goose is growing some baked in features to help with that - ie MCP servers are checked for malware (in the background), there is a feature to check for malicious prompts and future stuff to do with prompt injection and more (ie try to make security more ambient/in the background). Of course you can have it ask for permission for each step it takes etc. one other thing which may help (not always of interest) is being in rust, it is a monolithic binary that can be embedded in things pretty easily (of course if you use extensions, then they need their own runtimes available - so tradeoffs there) but that seems still a little unusual. hope that helps! |
Beta Was this translation helpful? Give feedback.
-
|
Thanks very much for taking time and providing detail update, is there any
link to read the architecture diagram and communication model
Also was interested to understand the underlying agentic framework it’s
built on and deployment of agents in each of the recipe executed say for
example code review
…On Tue, 18 Nov 2025 at 2:16 AM, Michael Neale ***@***.***> wrote:
we are trying to regularly run the "tbench" benchmark, which shows things
as competitive (it is not always easy to run it continuously), especially
with similar models (we also want to ensure it performs well with open
models). So performance I think can be similar (I think you will find many
things converge there, so then becomes preference based).
On security - some take an approach of asking permissions, fine grained
etc, goose has a few modes of operation from chat only, permission but my
favourite (and I expect most common) is the autonomous mode - which does
make you think differently about risk and security. goose is growing some
baked in features to help with that - ie MCP servers are checked for
malware (in the background), there is a feature to check for malicious
prompts and future stuff to do with prompt injection and more (ie try to
make security more ambient/in the background). Of course you can have it
ask for permission for each step it takes etc.
one other thing which may help (not always of interest) is being in rust,
it is a monolithic binary that can be embedded in things pretty easily (of
course if you use extensions, then they need their own runtimes available -
so tradeoffs there) but that seems still a little unusual.
hope that helps!
—
Reply to this email directly, view it on GitHub
<#3133 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AETH73YV23JPV2IECNUTE5L35IXZ5AVCNFSM6AAAAACAKDHENCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIOJZGQYDEMQ>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
I already have claude code, does this provide additional value?
Beta Was this translation helpful? Give feedback.
All reactions