agentmap is TS/JS-only. This issue is where that gets decided.
How to vote: react to a comment below with 👍. One comment per language. Reactions are counted; "+1" replies are not, because they can't be counted at a glance.
Read this before voting
A non-TS backend would not be the same product. agentmap's accuracy comes from ts-morph — a real TypeScript compiler with a type checker. Other languages would be parsed with tree-sitter, which reads syntax but has no type information and no module resolver.
That difference is not cosmetic. Here is what you would actually get:
| Query |
TS / JS today |
A non-TS language |
--relates (blast radius / who-imports) |
full |
full — the query that ports best |
--search (BM25 lexical) |
full |
full |
--hubs, --map, --symbols (PageRank) |
full |
full |
--find (where is X defined) |
full |
partial — no transitive re-export/barrel chains |
--any (unified router) |
full |
partial |
--callers / --calls (call graph) |
full |
none — it would refuse, explicitly |
--features / --feature |
Next.js App Router |
none |
| incremental rebuild |
yes |
no — full rebuild every time |
--callers and --calls are 259 lines of TypeScript language-service calls. Tree-sitter cannot reproduce them. The honest options are to refuse loudly or to guess by name-matching, and name-matching will not ship — a graph that silently mis-wires is worse than no graph, and it would falsify every accuracy claim this tool makes.
So when you vote, you're saying: yes, I'd use it even with the right three columns. If that's not true for you, say so — that's a more useful answer than a 👍, and it means you want a different tool than the one that could be built here.
Why a vote at all
The previous plan gated this on "wait until someone asks for Python." Nobody did — and the reason wasn't lack of interest. Every published version from 0.10.0 to 0.16.0 had a bug where the CLI printed nothing and exited 0 when run via npx or the npm bin. Anyone who tried it got silence, and left. Fixed in 0.16.1.
Meanwhile two people forked to add a language rather than open an issue:
If you're one of them: I'd rather upstream your work than compete with it. Please comment.
What happens with the votes
This is maintained by one person, part-time. Realistic capacity is two non-TS languages, ever — each one costs an ongoing maintenance tax forever, not just once. So this is a real ballot with a real cap, not a wishlist.
The bar to start work is deliberately about shape, not raw count: the leading language needs to be clearly ahead of second place, from at least 8 distinct accounts, with a majority explicitly accepting the table above.
If it doesn't clear that, the answer is a documented no for 12 months — which is a better outcome than a half-finished polyglot tool that does nothing well.
Full reasoning, including the arguments against doing this at all: ROADMAP.md, Part II.
You may have been sent here by agentmap itself — it counts your repo's files locally and prints this link when an unsupported language dominates. That count never leaves your machine; there is no telemetry in this tool and there will not be. Silence it with AGENTMAP_NO_CENSUS=1.
agentmap is TS/JS-only. This issue is where that gets decided.
How to vote: react to a comment below with 👍. One comment per language. Reactions are counted; "+1" replies are not, because they can't be counted at a glance.
Read this before voting
A non-TS backend would not be the same product. agentmap's accuracy comes from ts-morph — a real TypeScript compiler with a type checker. Other languages would be parsed with tree-sitter, which reads syntax but has no type information and no module resolver.
That difference is not cosmetic. Here is what you would actually get:
--relates(blast radius / who-imports)--search(BM25 lexical)--hubs,--map,--symbols(PageRank)--find(where is X defined)--any(unified router)--callers/--calls(call graph)--features/--feature--callersand--callsare 259 lines of TypeScript language-service calls. Tree-sitter cannot reproduce them. The honest options are to refuse loudly or to guess by name-matching, and name-matching will not ship — a graph that silently mis-wires is worse than no graph, and it would falsify every accuracy claim this tool makes.So when you vote, you're saying: yes, I'd use it even with the right three columns. If that's not true for you, say so — that's a more useful answer than a 👍, and it means you want a different tool than the one that could be built here.
Why a vote at all
The previous plan gated this on "wait until someone asks for Python." Nobody did — and the reason wasn't lack of interest. Every published version from 0.10.0 to 0.16.0 had a bug where the CLI printed nothing and exited 0 when run via
npxor the npm bin. Anyone who tried it got silence, and left. Fixed in 0.16.1.Meanwhile two people forked to add a language rather than open an issue:
rifanid98/agentmap-godstwn/agentmap-phpIf you're one of them: I'd rather upstream your work than compete with it. Please comment.
What happens with the votes
This is maintained by one person, part-time. Realistic capacity is two non-TS languages, ever — each one costs an ongoing maintenance tax forever, not just once. So this is a real ballot with a real cap, not a wishlist.
The bar to start work is deliberately about shape, not raw count: the leading language needs to be clearly ahead of second place, from at least 8 distinct accounts, with a majority explicitly accepting the table above.
If it doesn't clear that, the answer is a documented no for 12 months — which is a better outcome than a half-finished polyglot tool that does nothing well.
Full reasoning, including the arguments against doing this at all: ROADMAP.md, Part II.
You may have been sent here by agentmap itself — it counts your repo's files locally and prints this link when an unsupported language dominates. That count never leaves your machine; there is no telemetry in this tool and there will not be. Silence it with
AGENTMAP_NO_CENSUS=1.