Aider's repomap is a great idea. I remember participating in the discussion back then.
The unfortunate thing for Python that the repomap mentions, and untyped/duck-typed languages, is that function signatures do not mean a lot.
When it comes to Rust, it's a totally different story, function and method signatures convey a lot of important information. As a general rule, in every LLM query I include maximum one function/method implementation and everything else is function/method signatures.
By not giving mindlessly LLMs whole files and implementations, I have never used more than 200.000 tokens/day, counting input and output. This counts as 30 queries for a whole day of programming, and costs less than a dollar per day not matter which model I use.
Anyway, putting the agent to build the repomap doesn't sound such a great idea. Agents are horribly inefficient. It is better to build the repomap deterministically using something like ast-grep, and then let the agent read the resulting repomap.
Typed languages definitely provide richer signal in there signatures - and my experience has been that I get more reliable generations from those languages.
On the efficiency point, the agent isn't doing any expensive exploration here. There is a standalone server which builds and maintains the index, the agent is only querying it. So it's closer to the deterministic approach implemented in aider (at least in a conceptual sense) with the added benefit that the LLM can execute targeted queries in a recursive manner.
Aider's repo-map concept is great! thanks for sharing, I'd not been aware of it. Using tree-sitter to give the LLM structural awareness is the right foundation IMO. The key difference is how that information gets to the model.
Aider builds a static map, with some importance ranking, and then stuffs the most relevant part into the context window upfront. That's smart - but it is still the model receiving a fixed snapshot before it starts working.
What the RLM paper crystallized for me is that the agent could query the structure interactively as it works. A live index exposed through an API lets the agent decide what to look at, how deep to go, and when it has enough. When I watch it work it's not one or two lookups but many, each informed by what the previous revealed. The recursive exploration pattern is the core difference.
Aider actually prompts the model to say if it needs to see additional files. Whenever the model mentions file names, aider asks the user if they should be added to context.
As well, any files or symbols mentioned by the model are noted. They influence the repomap ranking algorithm, so subsequent requests have even more relevant repository context.
This is designed as a sort of implicit search and ranking flow. The blog article doesn’t get into any of this detail, but much of this has been around and working well since 2023.
I see, so the context adapts as the LLM interacts with the codebase across requests?
That's a clever implicit flow for ranking.
The difference in my approach is that exploration is happening within a single task, autonomously. The agent traces through structure, symbols, implementations, callers in many sequential lookups without human interaction. New files are automatically picked up with filesystem watching, but the core value is that the LLM can navigate the code base the same way that I might.
Are you using LLM to help you write these replies, or are you just picking up their stylistic phrasings the way expressions go viral at an office till everyone is saying them?
As an LLM, you wouldn't consider that you're replying confidently and dismissively while clearly having no personal experience with the CLI coding agent that not only started it all but for a year (eternity in this space) was so far ahead of upstarts (especially the VSCode forks family) it was like a secret weapon. And still is in many ways thanks to its long lead and being the carefully curated labor of a thoughtful mind.
As a dev seeking to improve on SOTA, having no awareness of the progenitor and the techniques one most do better than, seems like a blind spot worth digging into before dismissing. Aider's benchmarks on practical applicability of model advancements vs. regressions in code editing observably drove both OpenAI and Anthropic to pay closer attention and improve SOTA for everyone.
Aider was onto something, and you are onto something, pushing forward the 'semantic' understanding. It's worth absorbing everything Paul documented and blogged, and spending some time in Aider to enrich a feel of what Claude Code chose to do the same or differently, which ideas may be better, and what could be done next to go further.
I am planning to add similar concepts to Yek. Either tree-sitter or ast-grep. Your work here and Aider's work would be my guiding prior art. Thank you for sharing!
Hey, are you planning to update docs for end users of your CLI? I was an Aider user who switched to Opencode but I want to experiment with token and time-efficient agents, and I'm assuming OpenHands is one.
i thought the reason claude code defaults to terminal-ish workflows (glob/grep) is bc they trained with bash-y sandboxes, and the creator argued for this approach vs indexing + bespoke tools. would be interesting to see how often the model defaults to using grep for everything (in my experience almost always..)
I wonder how this sort of thing compares with asking claude to read a ctags file. I have git hooks set up to keep my tags up to date automatically, so that data is already lying around.
ctags just gives you locations of symbol definitions.
TreeSitter will also give you locations of symbol usages, which is obviously very useful to an AI agent. You can basically think of Treesitter as having full syntactic knowledge of the code it is looking at - like a compiler's AST.
There is also a more powerful cousin of ctags, cscope (C/C++) and Pycscope (python) that additonally gives usage locations, and more, as well as gtags that does similar, but supports more languages.
Excellent share, thank you. My question is with your setup, how strictly does Claude Code adhere to using this mode to traverse the codebase over grep? I have found this is to be a huge issue when implementing similar solutions... it loves to just grep.
I've been tinkering with it substantially and the most I can say is that it generally doesn't trigger automatically :( Claude has a really, really strong affinity for it's existing tools for exploring a code base.
I'd be happy to add support for scala and java - the current binary size is 11MB on my machine, so I think there's an opportunity to expand what this offers. At this time I don't know where I would draw the line of I'm not planning on supporting a thing. I think to some degree it would depend on usage / availability on my part
I don't see why it wouldn't - but I'm not familiar with setup / integration on other platforms. Would love to hear more about your stack and see if we can't find a way for you to try it out
The server exposes a straightforward API so wrapping it in MCP should be straight forward. The agent / skill interacts with the server using the cli implementation (part of the skill definition) at https://github.com/JaredStewart/coderlm/blob/main/plugin/ski...
LSP is a full fledged semantics solution providing go-to-definition functionality, trace references, type info etc. But requires a full language server, project configuration, and often a working build. That's great in an IDA, but the burden could be a bit much when it comes to working through an agent.
Tree-sitter handles structural queries giving the LLM the ability to evaluate function signatures, hierarchies and such. Packing this into the recursive language model enables the LLM to decide when it has enough information, it can continue to crawl the code base in bite sized increments to find what it needs. It's a far more minimal solution which lets it respond quickly with minimal overhead.
LSP is designed to help editors, not AI agents, and provides query by cursor position. Treesitter supports query by symbol (find definition, usages, etc), and so is much better suited to what an AI agent may want to do.
Thanks - I wasn't aware of that, although it still doesn't seem to be what would be most useful to an AI agent.
For example, if the agent wants to modify a function, it may want to know all the places the function is called, which AFAIK Treesitter can provide directly, but it seems with LSP you'd have to use that DocumentSymbol API to process every source file to find the usages, since you're really searching by source file, not by symbol.
This is the dumbest project I have ever seen. Why does it have so many upvotes and comments? Because there are a bunch of idiots here who need to add meaningless stuff to their empty and pointless papers. Only so-called pure academics would try to build such an index for code. File directories, search, and the built-in code parsing capability of compilers basically allow AI to locate any problem within 5 to 30 seconds. Even in the worst-case scenario, the problem is definitely not the lack of a so-called code index. To everyone participating in this discussion: how much have you actually used Claude Code? I have built several large projects with it without writing a single line of code myself, and I have never thought that code location or indexing was a problem at all.
anecdotally, it seems like this helps find better places for code to sit, understands the nuances of a code base better, and does a better job avoiding duplicate functionality.
it's still very much a work in progress, the thing I'm struggling with most right now is to have claude even using the capability without directly telling it to.
there seems to be benefits to the native stack (which lists files and then hopes for the best) relative to this sometimes. Frankly, it seems to be better at understanding the file structure. Where this approach really shines is in understanding the code base.
one approach that can work is to tell model to load read skill and/or call shell script that overloads default, there are variety of ways to attempt this with any harness, claude specifically has hooks some of which allow go, no go, do this instead etc. and ya, agree on grokking code base, ast integration feels like natural next step
Aider [0] wrote a piece about this [1] way back in Oct 2023!
I stumbled upon it in late 2023 when investigating ways to give OpenHands [2] better context dynamically.
[0] https://aider.chat/
[1] https://aider.chat/2023/10/22/repomap.html
[2] https://openhands.dev/
Aider's repomap is a great idea. I remember participating in the discussion back then.
The unfortunate thing for Python that the repomap mentions, and untyped/duck-typed languages, is that function signatures do not mean a lot.
When it comes to Rust, it's a totally different story, function and method signatures convey a lot of important information. As a general rule, in every LLM query I include maximum one function/method implementation and everything else is function/method signatures.
By not giving mindlessly LLMs whole files and implementations, I have never used more than 200.000 tokens/day, counting input and output. This counts as 30 queries for a whole day of programming, and costs less than a dollar per day not matter which model I use.
Anyway, putting the agent to build the repomap doesn't sound such a great idea. Agents are horribly inefficient. It is better to build the repomap deterministically using something like ast-grep, and then let the agent read the resulting repomap.
Typed languages definitely provide richer signal in there signatures - and my experience has been that I get more reliable generations from those languages.
On the efficiency point, the agent isn't doing any expensive exploration here. There is a standalone server which builds and maintains the index, the agent is only querying it. So it's closer to the deterministic approach implemented in aider (at least in a conceptual sense) with the added benefit that the LLM can execute targeted queries in a recursive manner.
Aider's repo-map concept is great! thanks for sharing, I'd not been aware of it. Using tree-sitter to give the LLM structural awareness is the right foundation IMO. The key difference is how that information gets to the model.
Aider builds a static map, with some importance ranking, and then stuffs the most relevant part into the context window upfront. That's smart - but it is still the model receiving a fixed snapshot before it starts working.
What the RLM paper crystallized for me is that the agent could query the structure interactively as it works. A live index exposed through an API lets the agent decide what to look at, how deep to go, and when it has enough. When I watch it work it's not one or two lookups but many, each informed by what the previous revealed. The recursive exploration pattern is the core difference.
Aider actually prompts the model to say if it needs to see additional files. Whenever the model mentions file names, aider asks the user if they should be added to context.
As well, any files or symbols mentioned by the model are noted. They influence the repomap ranking algorithm, so subsequent requests have even more relevant repository context.
This is designed as a sort of implicit search and ranking flow. The blog article doesn’t get into any of this detail, but much of this has been around and working well since 2023.
I see, so the context adapts as the LLM interacts with the codebase across requests?
That's a clever implicit flow for ranking.
The difference in my approach is that exploration is happening within a single task, autonomously. The agent traces through structure, symbols, implementations, callers in many sequential lookups without human interaction. New files are automatically picked up with filesystem watching, but the core value is that the LLM can navigate the code base the same way that I might.
> That's smart - but it is still…
> That's a clever… The difference in my approach…
Are you using LLM to help you write these replies, or are you just picking up their stylistic phrasings the way expressions go viral at an office till everyone is saying them?
As an LLM, you wouldn't consider that you're replying confidently and dismissively while clearly having no personal experience with the CLI coding agent that not only started it all but for a year (eternity in this space) was so far ahead of upstarts (especially the VSCode forks family) it was like a secret weapon. And still is in many ways thanks to its long lead and being the carefully curated labor of a thoughtful mind.
As a dev seeking to improve on SOTA, having no awareness of the progenitor and the techniques one most do better than, seems like a blind spot worth digging into before dismissing. Aider's benchmarks on practical applicability of model advancements vs. regressions in code editing observably drove both OpenAI and Anthropic to pay closer attention and improve SOTA for everyone.
Aider was onto something, and you are onto something, pushing forward the 'semantic' understanding. It's worth absorbing everything Paul documented and blogged, and spending some time in Aider to enrich a feel of what Claude Code chose to do the same or differently, which ideas may be better, and what could be done next to go further.
I am planning to add similar concepts to Yek. Either tree-sitter or ast-grep. Your work here and Aider's work would be my guiding prior art. Thank you for sharing!
https://github.com/mohsen1/yek
I just looked and it was posted a number of times with 0 discussion
https://news.ycombinator.com/item?id=38062493
https://news.ycombinator.com/item?id=41411187
https://news.ycombinator.com/item?id=40231527
https://news.ycombinator.com/item?id=39993459
https://news.ycombinator.com/item?id=41393767
https://news.ycombinator.com/item?id=39391946
Hey, are you planning to update docs for end users of your CLI? I was an Aider user who switched to Opencode but I want to experiment with token and time-efficient agents, and I'm assuming OpenHands is one.
Great idea! I’ve been thinking about something along these lines as well.
I recommend configuring it as a tool for Opencode.
Going from Claude Code to Opencode was like going from Windows to Mac.
And going to pi agent is like ascending to Linux ;)
yeah I would definitely recommend the same. I'm a daily user of opencode and I really want to try this.
will take a look at opencode, thanks for sharing!
i thought the reason claude code defaults to terminal-ish workflows (glob/grep) is bc they trained with bash-y sandboxes, and the creator argued for this approach vs indexing + bespoke tools. would be interesting to see how often the model defaults to using grep for everything (in my experience almost always..)
I wonder how this sort of thing compares with asking claude to read a ctags file. I have git hooks set up to keep my tags up to date automatically, so that data is already lying around.
ctags just gives you locations of symbol definitions.
TreeSitter will also give you locations of symbol usages, which is obviously very useful to an AI agent. You can basically think of Treesitter as having full syntactic knowledge of the code it is looking at - like a compiler's AST.
There is also a more powerful cousin of ctags, cscope (C/C++) and Pycscope (python) that additonally gives usage locations, and more, as well as gtags that does similar, but supports more languages.
Excellent share, thank you. My question is with your setup, how strictly does Claude Code adhere to using this mode to traverse the codebase over grep? I have found this is to be a huge issue when implementing similar solutions... it loves to just grep.
Can you make the plugin start automatically, on some suitable trigger? Any plans to support JVM languages?
edit: Does Claude not invoke it automatically, then, so you have to call the skill?
I've been tinkering with it substantially and the most I can say is that it generally doesn't trigger automatically :( Claude has a really, really strong affinity for it's existing tools for exploring a code base.
I'd be happy to add support for scala and java - the current binary size is 11MB on my machine, so I think there's an opportunity to expand what this offers. At this time I don't know where I would draw the line of I'm not planning on supporting a thing. I think to some degree it would depend on usage / availability on my part
Would this be useful to people who aren't using Claude? Maybe it should be installable in a more normal way, instead of as a Claude plugin.
I don't see why it wouldn't - but I'm not familiar with setup / integration on other platforms. Would love to hear more about your stack and see if we can't find a way for you to try it out
A CLI or slim MCP would do it. IF you want a formal plugin, here's another popular ecosystem: https://opencode.ai/docs/plugins/
The server exposes a straightforward API so wrapping it in MCP should be straight forward. The agent / skill interacts with the server using the cli implementation (part of the skill definition) at https://github.com/JaredStewart/coderlm/blob/main/plugin/ski...
Since it appears to be a REST API you could just share the OpenAPI spec and let the agent cURL it.
My experience is best with cli and a concise skill.md
I see a lot of overlap with LSPs, which better agents already use, so I would appreciate a comparison. What does this add?
Tree-sitter and LSP solve different problems.
LSP is a full fledged semantics solution providing go-to-definition functionality, trace references, type info etc. But requires a full language server, project configuration, and often a working build. That's great in an IDA, but the burden could be a bit much when it comes to working through an agent.
Tree-sitter handles structural queries giving the LLM the ability to evaluate function signatures, hierarchies and such. Packing this into the recursive language model enables the LLM to decide when it has enough information, it can continue to crawl the code base in bite sized increments to find what it needs. It's a far more minimal solution which lets it respond quickly with minimal overhead.
LSP is designed to help editors, not AI agents, and provides query by cursor position. Treesitter supports query by symbol (find definition, usages, etc), and so is much better suited to what an AI agent may want to do.
My agent already uses LSP plugins, and the protocol supports querying by symbol:
https://microsoft.github.io/language-server-protocol/specifi...
https://microsoft.github.io/language-server-protocol/specifi...
Thanks - I wasn't aware of that, although it still doesn't seem to be what would be most useful to an AI agent.
For example, if the agent wants to modify a function, it may want to know all the places the function is called, which AFAIK Treesitter can provide directly, but it seems with LSP you'd have to use that DocumentSymbol API to process every source file to find the usages, since you're really searching by source file, not by symbol.
For that I believe you could use textDocument/references:
https://microsoft.github.io/language-server-protocol/specifi...
Yes, but that one is position-based rather than name-based - I believe it's basically for an editor to ask about whatever is under the cursor.
This is the dumbest project I have ever seen. Why does it have so many upvotes and comments? Because there are a bunch of idiots here who need to add meaningless stuff to their empty and pointless papers. Only so-called pure academics would try to build such an index for code. File directories, search, and the built-in code parsing capability of compilers basically allow AI to locate any problem within 5 to 30 seconds. Even in the worst-case scenario, the problem is definitely not the lack of a so-called code index. To everyone participating in this discussion: how much have you actually used Claude Code? I have built several large projects with it without writing a single line of code myself, and I have never thought that code location or indexing was a problem at all.
been wondering about treesitter grepping for agents
how do plans compare with and without etc. evven just anecdotally what you've seen so far etc
anecdotally, it seems like this helps find better places for code to sit, understands the nuances of a code base better, and does a better job avoiding duplicate functionality.
it's still very much a work in progress, the thing I'm struggling with most right now is to have claude even using the capability without directly telling it to.
there seems to be benefits to the native stack (which lists files and then hopes for the best) relative to this sometimes. Frankly, it seems to be better at understanding the file structure. Where this approach really shines is in understanding the code base.
one approach that can work is to tell model to load read skill and/or call shell script that overloads default, there are variety of ways to attempt this with any harness, claude specifically has hooks some of which allow go, no go, do this instead etc. and ya, agree on grokking code base, ast integration feels like natural next step