Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-20 01:31:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	ggml/src/ggml-cpu/CMakeLists.txt
#	ggml/src/ggml-cpu/kleidiai/kernels.cpp
#	ggml/src/ggml-cpu/kleidiai/kernels.h
#	ggml/src/ggml-cpu/kleidiai/kleidiai.cpp
#	ggml/src/ggml-cpu/ops.cpp
#	ggml/src/ggml-cuda/mmq.cuh
#	ggml/src/ggml-hexagon/htp/hmx-queue.c
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_iq4_nl_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q1_0_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_0_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_0_f32_spec.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_1_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_k_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_0_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_1_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_k_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q6_k_f32.cl
#	ggml/src/ggml-opencl/kernels/gemv_noshuffle_q8_0_f32.cl
#	ggml/src/ggml-opencl/kernels/mul_mv_f16_f16.cl
#	ggml/src/ggml-opencl/kernels/mul_mv_f16_f32.cl
#	ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_1row.cl
#	ggml/src/ggml-sycl/fattn-vec.hpp
#	tests/test-backend-ops.cpp
#	tests/test-chat-auto-parser.cpp
#	tests/test-export-graph-ops.cpp
#	tests/test-jinja.cpp
#	tests/test-llama-archs.cpp
#	tools/tokenize/tokenize.cpp
This commit is contained in:
Concedo
2026-07-15 15:39:31 +08:00
parent 23ba6932b5 12127defda
commit 9001369da0
73 changed files with 6859 additions and 3973 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/server/server-context.cpp
+33 -38
View File
@@ -220,8 +220,6 @@ struct server_slot {
return false;
}
GGML_ASSERT(prompt.data.size() == 0);
const size_t cur_size_tgt = llama_state_seq_get_size_ext(ctx_tgt, id, LLAMA_STATE_SEQ_FLAGS_NONE);
const size_t cur_size_dft = ctx_dft ? llama_state_seq_get_size_ext(ctx_dft, id, LLAMA_STATE_SEQ_FLAGS_NONE) : 0;
@@ -252,11 +250,7 @@ struct server_slot {
return res;
}
void prompt_clear(bool allow_processing) {
if (!allow_processing) {
GGML_ASSERT(!is_processing());
}
void prompt_clear() {
SLT_TRC(*this, "clearing prompt with %zu tokens\n", prompt.tokens.size());
common_context_seq_rm(ctx_tgt, id, -1, -1);
@@ -264,7 +258,7 @@ struct server_slot {
common_context_seq_rm(ctx_dft, id, -1, -1);
}
prompt.tokens.clear();
prompt.clear();
}
std::vector<common_adapter_lora_info> lora;
@@ -493,7 +487,7 @@ struct server_slot {
// do not keep context of the child slots - the parent's context is enough
if (task->is_child()) {
prompt_clear(false);
prompt_clear();
}
reset();
@@ -1626,7 +1620,7 @@ private:
ret->prompt_save(*prompt_cache);
if (!ret->prompt_load(*prompt_cache, task.tokens)) {
ret->prompt_clear(false);
ret->prompt_clear();
}
prompt_cache->update();
@@ -1658,7 +1652,7 @@ private:
if (slot.prompt.n_tokens() > 0) {
SRV_WRN("purging slot %d with %zu tokens\n", slot.id, slot.prompt.tokens.size());
slot.prompt_clear(false);
slot.prompt_clear();
res = true;
@@ -1691,7 +1685,7 @@ private:
// if lora has changed, check to see if the cache should be cleared
if (lora_should_clear_cache(slot.lora, task_loras)) {
SLT_TRC(slot, "clearing cache for lora change. %zu loras -> %zu loras\n", slot.lora.size(), task.params.lora.size());
slot.prompt.tokens.clear();
slot.prompt.clear();
} else {
SLT_TRC(slot, "keeping cache for alora. %zu target loras\n", task_loras.size());
}
@@ -2405,7 +2399,7 @@ private:
if (params_base.kv_unified) {
// [TAG_IDLE_SLOT_CLEAR]
slot.prompt_clear(false);
slot.prompt_clear();
}
}
}
@@ -2573,12 +2567,12 @@ private:
size_t token_count = 0;
size_t nread = llama_state_seq_load_file(ctx_tgt, filepath.c_str(), slot->id, tokens.data(), tokens.size(), &token_count);
if (nread == 0) {
slot->prompt.tokens.clear(); // KV may already been invalidated?
slot->prompt.clear(); // KV may already been invalidated?
send_error(task, "Unable to restore slot, no available space in KV cache or invalid slot save file", ERROR_TYPE_INVALID_REQUEST);
break;
}
tokens.resize(token_count);
slot->prompt.tokens.clear();
slot->prompt.clear();
slot->prompt.tokens.insert(tokens);
const int64_t t_end = ggml_time_us();
@@ -2615,7 +2609,7 @@ private:
// Erase token cache
const size_t n_erased = slot->prompt.tokens.size();
slot->prompt_clear(false);
slot->prompt_clear();
auto res = std::make_unique<server_task_result_slot_erase>();
res->id = task.id;
@@ -2775,6 +2769,27 @@ private:
abort_all_slots("pre_decode() failed: " + std::string(e.what()));
}
GGML_ASSERT(batch.slot_batched || batch.size() == 0);
if (batch.slot_batched) {
auto & slot_batched = batch.slot_batched;
auto & alora_scale = batch.alora_scale;
auto & alora_disabled_id = batch.alora_disabled_id;
// TODO @ngxson : alora handling is too messy, need to refactor it to be more clear and maintainable
// apply lora, only need to do it once per batch
common_set_adapter_lora(ctx_tgt, slot_batched->lora);
// if the lora is temporarily disabled for an alora, re-enable it
// for next time
if (alora_scale > 0.0f) {
SRV_DBG("re-enabling alora with scale %f\n", alora_scale);
slot_batched->lora[alora_disabled_id].scale = alora_scale;
}
llama_set_embeddings(ctx_tgt, slot_batched->need_embd());
}
llama_batch batch_view;
int32_t off_next = 0;
int32_t n_batch = llama_n_batch(ctx_tgt);
@@ -2814,7 +2829,6 @@ private:
abort_all_slots("post_decode() failed: " + std::string(e.what()));
break; // stop any further processing
}
}
}
@@ -2880,7 +2894,7 @@ private:
new_tokens.resize(slot.prompt.tokens.size() - n_discard);
slot.prompt.tokens.clear();
slot.prompt.clear();
slot.prompt.tokens.insert(new_tokens);
}
@@ -3556,25 +3570,6 @@ private:
bool decode(int32_t & n_batch, int32_t off, llama_batch & batch_view) {
SRV_DBG("n_batch (effective) = %d, off = %d\n", n_batch, off);
auto & slot_batched = batch.slot_batched;
auto & alora_scale = batch.alora_scale;
auto & alora_disabled_id = batch.alora_disabled_id;
// TODO @ngxson : alora handling is too messy, need to refactor it to be more clear and maintainable
if (slot_batched) {
// apply lora, only need to do it once per batch
common_set_adapter_lora(ctx_tgt, slot_batched->lora);
// if the lora is temporarily disabled for an alora, re-enable it
// for next time
if (alora_scale > 0.0f) {
SRV_DBG("re-enabling alora with scale %f\n", alora_scale);
slot_batched->lora[alora_disabled_id].scale = alora_scale;
}
llama_set_embeddings(ctx_tgt, slot_batched->need_embd());
}
if (batch.size() == 0) {
SRV_WRN("%s", "no tokens to decode\n");
@@ -3622,7 +3617,7 @@ private:
// note: it's complicated to keep track of how much of the current batch has been
// processed before the error occurred, so we simply clear the entire context
slot.prompt_clear(false);
slot.prompt_clear();
}
}
tools/server/server-http.cpp
+28 -5
View File
@@ -47,6 +47,16 @@ static void log_server_request(const httplib::Request & req, const httplib::Resp
SRV_DBG("response: %s\n", res.body.c_str());
}
// returns true if the Origin header value's host is localhost / 127.0.0.1 / ::1 (any port)
static bool origin_is_localhost(const std::string & origin) {
try {
const std::string host = common_http_parse_url(origin).host;
return host == "localhost" || host == "127.0.0.1" || host == "::1";
} catch (const std::exception &) {
return false;
}
}
// For Google Cloud Platform deployment compatibility
struct gcp_params {
bool enabled;
@@ -266,13 +276,26 @@ bool server_http_context::init(const common_params & params) {
};
// register server middlewares
srv->set_pre_routing_handler([middleware_validate_api_key, middleware_server_state](const httplib::Request & req, httplib::Response & res) {
res.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin"));
srv->set_pre_routing_handler([&params, middleware_validate_api_key, middleware_server_state](const httplib::Request & req, httplib::Response & res) {
if (params.cors_credentials && params.cors_origins == "*") {
// special case: echo back the Origin header to allow any origin to access the server with credentials
res.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin"));
} else if (params.cors_origins == "localhost") {
// special case: only reflect the Origin header if it is a localhost origin
std::string origin = req.get_header_value("Origin");
if (origin_is_localhost(origin)) {
res.set_header("Access-Control-Allow-Origin", origin);
} else {
SRV_WRN("(CORS) skip non-localhost origin: %s\n", origin.c_str());
}
} else {
res.set_header("Access-Control-Allow-Origin", params.cors_origins);
}
// If this is OPTIONS request, skip validation because browsers don't include Authorization header
if (req.method == "OPTIONS") {
res.set_header("Access-Control-Allow-Credentials", "true");
res.set_header("Access-Control-Allow-Methods", "GET, POST");
res.set_header("Access-Control-Allow-Headers", "*");
res.set_header("Access-Control-Allow-Credentials", params.cors_credentials ? "true" : "false");
res.set_header("Access-Control-Allow-Methods", params.cors_methods);
res.set_header("Access-Control-Allow-Headers", params.cors_headers);
res.set_content("", "text/html"); // blank response, no data
return httplib::Server::HandlerResponse::Handled; // skip further processing
}
tools/server/server-task.cpp
+14 -12
View File
@@ -1646,16 +1646,16 @@ size_t server_prompt_cache::n_tokens() const {
size_t res = 0;
for (const auto & state : states) {
res += state.n_tokens();
res += state.prompt.n_tokens();
}
return res;
}
server_prompt * server_prompt_cache::alloc(const server_prompt & prompt, size_t state_size_tgt, size_t state_size_dft) {
server_prompt_cache_state * server_prompt_cache::alloc(const server_prompt & prompt, size_t state_size_tgt, size_t state_size_dft) {
// first check if the current state is contained fully in the cache
for (auto it = states.begin(); it != states.end(); ++it) {
const int cur_lcp_len = it->tokens.get_common_prefix(prompt.tokens);
const int cur_lcp_len = it->prompt.tokens.get_common_prefix(prompt.tokens);
if (cur_lcp_len == (int) prompt.tokens.size()) {
SRV_TRC("%s", " - prompt is already in the cache, skipping\n");
@@ -1680,9 +1680,9 @@ server_prompt * server_prompt_cache::alloc(const server_prompt & prompt, size_t
// remove any cached prompts that are fully contained in the current prompt
for (auto it = states.begin(); it != states.end();) {
const int len = it->tokens.get_common_prefix(prompt.tokens);
const int len = it->prompt.tokens.get_common_prefix(prompt.tokens);
if (len == (int) it->tokens.size()) {
if (len == (int) it->prompt.tokens.size()) {
SRV_TRC(" - removing obsolete cached prompt with length %d\n", len);
it = states.erase(it);
@@ -1721,12 +1721,14 @@ server_prompt * server_prompt_cache::alloc(const server_prompt & prompt, size_t
}
states.push_back({
/*.tokens =*/ prompt.tokens.clone(),
/*.data =*/ {
/*.prompt =*/ {
/*.tokens =*/ prompt.tokens.clone(),
/*.checkpoints =*/ prompt.checkpoints,
},
/*.data =*/ {
/*.main =*/ std::move(state_data_tgt),
/*.drft =*/ std::move(state_data_dft),
},
/*.checkpoints =*/ prompt.checkpoints,
});
return &states.back();
@@ -1744,9 +1746,9 @@ bool server_prompt_cache::load(server_prompt & prompt, const server_tokens & tok
// find the most similar cached prompt, that would also preserve the most context
for (auto it = states.begin(); it != states.end(); ++it) {
const int lcp_cur = it->tokens.get_common_prefix(tokens_new);
const int lcp_cur = it->prompt.tokens.get_common_prefix(tokens_new);
const float f_keep_cur = float(lcp_cur) / it->tokens.size();
const float f_keep_cur = float(lcp_cur) / it->prompt.tokens.size();
const float sim_cur = float(lcp_cur) / tokens_new.size();
// don't trash large prompts
@@ -1799,7 +1801,7 @@ bool server_prompt_cache::load(server_prompt & prompt, const server_tokens & tok
}
}
prompt = std::move(*it_best);
prompt = std::move(it_best->prompt);
states.erase(it_best);
}
@@ -1836,6 +1838,6 @@ void server_prompt_cache::update() {
for (const auto & state : states) {
SRV_TRC(" - prompt %p: %7d tokens, checkpoints: %2zu, %9.3f MiB\n",
(const void *)&state, state.n_tokens(), state.checkpoints.size(), state.size() / (1024.0 * 1024.0));
(const void *)&state, state.prompt.n_tokens(), state.prompt.checkpoints.size(), state.size() / (1024.0 * 1024.0));
}
}
tools/server/server-task.h
+29 -24
View File
@@ -584,32 +584,14 @@ struct server_task_result_apply_lora : server_task_result {
virtual json to_json() override;
};
struct server_prompt_data {
std::vector<uint8_t> main;
std::vector<uint8_t> drft;
size_t size() const {
return main.size() + drft.size();
}
};
struct server_prompt {
server_tokens tokens;
server_prompt_data data;
std::list<common_prompt_checkpoint> checkpoints;
size_t size() const {
size_t res = 0;
res += data.size();
for (const auto & ckpt : checkpoints) {
res += ckpt.size();
}
return res;
void clear() {
tokens.clear();
checkpoints.clear();
}
int n_tokens() const {
@@ -619,19 +601,42 @@ struct server_prompt {
server_prompt clone() const {
return server_prompt {
tokens.clone(),
data,
checkpoints,
};
}
};
struct server_prompt_data {
std::vector<uint8_t> main;
std::vector<uint8_t> drft;
size_t size() const {
return main.size() + drft.size();
}
};
struct server_prompt_cache_state {
server_prompt prompt;
server_prompt_data data;
size_t size() const {
size_t res = data.size();
for (const auto & ckpt : prompt.checkpoints) {
res += ckpt.size();
}
return res;
}
};
struct server_prompt_cache {
server_prompt_cache(int32_t limit_size_mib, size_t limit_tokens) {
this->limit_size = 1024ull*1024ull*(limit_size_mib < 0 ? 0 : limit_size_mib);
this->limit_tokens = limit_tokens;
}
std::list<server_prompt> states;
std::list<server_prompt_cache_state> states;
// in bytes, 0 = no limit
size_t limit_size = 0;
@@ -643,7 +648,7 @@ struct server_prompt_cache {
size_t n_tokens() const;
server_prompt * alloc(const server_prompt & prompt, size_t state_size_main, size_t state_size_drft);
server_prompt_cache_state * alloc(const server_prompt & prompt, size_t state_size_main, size_t state_size_drft);
bool load(server_prompt & prompt, const server_tokens & tokens_new, llama_context * ctx_main, llama_context * ctx_drft, int32_t id_slot);
tools/server/server.cpp
+25 -11
View File
@@ -303,14 +303,24 @@ int llama_server(common_params & params, int argc, char ** argv) {
return res;
};
if (params.cors_origins == "*" && params.api_keys.empty()) {
SRV_WRN("%s", "-----------------\n");
SRV_WRN("%s", "CORS is set to allow all origins ('*') and no API key is set\n");
SRV_WRN("%s", "this can be a security risk (cross-origin attacks)\n");
SRV_WRN("%s", "more info: https://github.com/ggml-org/llama.cpp/pull/25655\n");
SRV_WRN("%s", "-----------------\n");
}
// CORS proxy (EXPERIMENTAL, only used by the Web UI for MCP)
std::vector<std::string> warn_names;
if (is_router_server) {
warn_names.push_back("router mode");
}
if (params.ui_mcp_proxy) {
SRV_WRN("%s", "-----------------\n");
SRV_WRN("%s", "CORS proxy is enabled, do not expose server to untrusted environments\n");
SRV_WRN("%s", "This feature is EXPERIMENTAL and may be removed or changed in future versions\n");
SRV_WRN("%s", "-----------------\n");
ctx_http.get ("/cors-proxy", ex_wrapper(proxy_handler_get));
ctx_http.post("/cors-proxy", ex_wrapper(proxy_handler_post));
warn_names.push_back("MCP proxy (experimental)");
} else {
ctx_http.get ("/cors-proxy", ex_wrapper(res_403));
ctx_http.post("/cors-proxy", ex_wrapper(res_403));
@@ -324,17 +334,24 @@ int llama_server(common_params & params, int argc, char ** argv) {
SRV_ERR("tools setup failed: %s\n", e.what());
return 1;
}
SRV_WRN("%s", "-----------------\n");
SRV_WRN("%s", "Built-in tools are enabled, do not expose server to untrusted environments\n");
SRV_WRN("%s", "This feature is EXPERIMENTAL and may be changed in the future\n");
SRV_WRN("%s", "-----------------\n");
ctx_http.get ("/tools", ex_wrapper(tools.handle_get));
ctx_http.post("/tools", ex_wrapper(tools.handle_post));
warn_names.push_back("built-in tools (experimental)");
} else {
ctx_http.get ("/tools", ex_wrapper(res_403));
ctx_http.post("/tools", ex_wrapper(res_403));
}
if (warn_names.size() > 0) {
SRV_WRN("%s", "-----------------\n");
SRV_WRN("%s", "the following feature(s) are enabled:\n");
for (const auto & name : warn_names) {
SRV_WRN(" %s\n", name.c_str());
}
SRV_WRN("%s", "do not expose the server to untrusted environments\n");
SRV_WRN("%s", "-----------------\n");
}
//
// Handle downloading model
//
@@ -452,9 +469,6 @@ int llama_server(common_params & params, int argc, char ** argv) {
SRV_INF("listening on %s\n", ctx_http.listening_address.c_str());
if (is_router_server) {
SRV_WRN("%s", "NOTE: router mode is experimental\n");
SRV_WRN("%s", " it is not recommended to use this mode in untrusted environments\n");
if (!params.models_preset_hf.empty()) {
SRV_WRN( "NOTE: using preset.ini from HF repo '%s'\n", params.models_preset_hf.c_str());
SRV_WRN("%s", " please only use presets that you can trust! Unknown presets may be unsafe\n");
tools/server/tests/unit/test_security.py
+65 -1
View File
@@ -91,7 +91,7 @@ def test_openai_library_correct_api_key():
("localhost", "Access-Control-Allow-Origin", "localhost"),
("web.mydomain.fr", "Access-Control-Allow-Origin", "web.mydomain.fr"),
("origin", "Access-Control-Allow-Credentials", "true"),
("web.mydomain.fr", "Access-Control-Allow-Methods", "GET, POST"),
("web.mydomain.fr", "Access-Control-Allow-Methods", "GET, POST, DELETE, OPTIONS"),
("web.mydomain.fr", "Access-Control-Allow-Headers", "*"),
])
def test_cors_options(origin: str, cors_header: str, cors_header_value: str):
@@ -107,6 +107,70 @@ def test_cors_options(origin: str, cors_header: str, cors_header_value: str):
assert res.headers[cors_header] == cors_header_value
@pytest.mark.parametrize("origin", [
"http://localhost",
"http://localhost:8080",
"http://127.0.0.1",
"http://127.0.0.1:3000",
"http://[::1]",
"http://[::1]:3000",
])
def test_cors_origins_localhost_reflects(origin: str):
global server
server = ServerPreset.router()
server.cors_origins = "localhost"
server.start()
res = server.make_request("OPTIONS", "/completions", headers={
"Origin": origin,
"Access-Control-Request-Method": "POST",
"Access-Control-Request-Headers": "Authorization",
})
assert res.status_code == 200
assert res.headers["Access-Control-Allow-Origin"] == origin
@pytest.mark.parametrize("origin", [
"http://web.mydomain.fr",
"http://evil.com",
"http://notlocalhost",
"http://localhost.evil.com",
])
def test_cors_origins_localhost_rejects(origin: str):
global server
server = ServerPreset.router()
server.cors_origins = "localhost"
server.start()
res = server.make_request("OPTIONS", "/completions", headers={
"Origin": origin,
"Access-Control-Request-Method": "POST",
"Access-Control-Request-Headers": "Authorization",
})
assert res.status_code == 200
assert "Access-Control-Allow-Origin" not in res.headers
def test_cors_origins_defaults_to_localhost_with_tools_enabled():
global server
server = ServerPreset.router()
server.server_tools = "all"
server.start()
res = server.make_request("OPTIONS", "/completions", headers={
"Origin": "http://localhost:8080",
"Access-Control-Request-Method": "POST",
"Access-Control-Request-Headers": "Authorization",
})
assert res.status_code == 200
assert res.headers["Access-Control-Allow-Origin"] == "http://localhost:8080"
res = server.make_request("OPTIONS", "/completions", headers={
"Origin": "http://evil.com",
"Access-Control-Request-Method": "POST",
"Access-Control-Request-Headers": "Authorization",
})
assert res.status_code == 200
assert "Access-Control-Allow-Origin" not in res.headers
def test_cors_proxy_only_forwards_explicit_proxy_headers():
class CaptureHeadersHandler(BaseHTTPRequestHandler):
def do_GET(self):
tools/server/tests/utils.py
+4 -1
View File
@@ -114,6 +114,7 @@ class ServerProcess:
backend_sampling: bool = False
gcp_compat: bool = False
server_tools: str | None = None
cors_origins: str | None = None
# session variables
process: subprocess.Popen | None = None
@@ -170,6 +171,8 @@ class ServerProcess:
server_args.extend(["--models-max", self.models_max])
if self.models_preset:
server_args.extend(["--models-preset", self.models_preset])
if self.cors_origins:
server_args.extend(["--cors-origins", self.cors_origins])
if self.n_batch:
server_args.extend(["--batch-size", self.n_batch])
if self.n_ubatch:
@@ -359,7 +362,7 @@ class ServerProcess:
if parse_body:
try:
result.body = response.json()
except JSONDecodeError:
except (JSONDecodeError, requests.exceptions.JSONDecodeError):
result.body = response.text
else:
result.body = None
tools/ui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActionAdd/ChatFormActionAddMcpServersSubmenu.svelte
+4 -4
View File
@@ -17,10 +17,10 @@
let { onMcpSettingsClick }: Props = $props();
let mcpSearchQuery = $state('');
let allMcpServers = $derived(mcpStore.getServers());
let mcpServers = $derived(mcpStore.visibleMcpServers);
// Every configured server is listed; `enabled` is an on/off state,
// not a visibility filter, so a disabled server stays toggleable.
let mcpServers = $derived(mcpStore.getServers());
let hasMcpServers = $derived(mcpServers.length > 0);
// let hasAnyMcpServers = $derived(allMcpServers.length > 0);
let filteredMcpServers = $derived.by(() => {
const query = mcpSearchQuery.toLowerCase().trim();
if (!query) return mcpServers;
@@ -46,7 +46,7 @@
function handleMcpSubMenuOpen(open: boolean) {
if (open) {
mcpSearchQuery = '';
mcpStore.runHealthChecksForServers(allMcpServers);
mcpStore.runHealthChecksForServers(mcpServers);
}
}
tools/ui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActionAdd/ChatFormActionAddReasoningSubmenu.svelte
+20 -72
View File
@@ -2,85 +2,31 @@
import { Lightbulb, LightbulbOff, Check, Info } from '@lucide/svelte';
import * as DropdownMenu from '$lib/components/ui/dropdown-menu';
import * as Tooltip from '$lib/components/ui/tooltip';
import { ReasoningEffort } from '$lib/enums';
import { REASONING_EFFORT_TOKENS } from '$lib/constants/reasoning-effort-tokens';
import { REASONING_EFFORT_LEVELS } from '$lib/constants/reasoning-effort';
import type { ReasoningEffortLevel } from '$lib/types';
import {
modelsStore,
checkModelSupportsThinking,
supportsThinking,
propsCacheVersion,
loadedModelIds
} from '$lib/stores/models.svelte';
import { chatStore } from '$lib/stores/chat.svelte';
import { conversationsStore, activeMessages } from '$lib/stores/conversations.svelte';
import { isRouterMode } from '$lib/stores/server.svelte';
import type { DatabaseMessage } from '$lib/types/database';
import { useReasoningMenu } from '$lib/hooks/use-reasoning-menu.svelte';
let subOpen = $state(false);
let conversationModel = $derived(
chatStore.getConversationModel(activeMessages() as DatabaseMessage[])
);
let modelSupportsThinkingFromMessages = $derived.by(() => {
const modelId = isRouterMode() ? modelsStore.selectedModelName || conversationModel : null;
if (!modelId) return false;
const messages = conversationsStore.activeMessages;
return messages.some(
(m) => m.role === 'assistant' && m.model === modelId && !!m.reasoningContent
);
});
let modelSupportsThinking = $derived.by(() => {
loadedModelIds();
propsCacheVersion();
if (isRouterMode()) {
const modelId = modelsStore.selectedModelName || conversationModel;
return checkModelSupportsThinking(modelId ?? '') || modelSupportsThinkingFromMessages;
}
return supportsThinking() || modelSupportsThinkingFromMessages;
});
let thinkingEnabled = $derived(conversationsStore.getThinkingEnabled());
let currentEffort = $derived(conversationsStore.getReasoningEffort());
let isOff = $derived(!thinkingEnabled);
function isSelected(item: ReasoningEffortLevel): boolean {
if (item.isOff) return isOff;
return thinkingEnabled && currentEffort === item.value;
}
function handleSelection(item: ReasoningEffortLevel) {
if (item.isOff) {
conversationsStore.setThinkingEnabled(false);
} else {
conversationsStore.setThinkingEnabled(true);
conversationsStore.setReasoningEffort(item.value as ReasoningEffort);
}
subOpen = false;
}
const reasoning = useReasoningMenu();
</script>
{#if modelSupportsThinking}
{#if reasoning.modelSupportsThinking}
<DropdownMenu.Sub bind:open={subOpen}>
<DropdownMenu.SubTrigger class="flex cursor-pointer items-center gap-2">
{#if thinkingEnabled}
{#if reasoning.thinkingEnabled}
<Lightbulb class="h-4 w-4 shrink-0 text-amber-400" />
{:else}
<LightbulbOff class="h-4 w-4 shrink-0 text-muted-foreground" />
{/if}
<span class="text-sm inline-flex gap-2 {!thinkingEnabled ? 'text-muted-foreground' : ''}">
<span
class="text-sm inline-flex gap-2 {!reasoning.thinkingEnabled
? 'text-muted-foreground'
: ''}"
>
Reasoning
<span class="capitalize text-muted-foreground">
{thinkingEnabled ? currentEffort : 'off'}
{reasoning.thinkingEnabled ? reasoning.currentEffort : 'off'}
</span>
</span>
</DropdownMenu.SubTrigger>
@@ -88,14 +34,18 @@
<DropdownMenu.SubContent
class="w-60 bg-popover p-1.5 text-popover-foreground shadow-md outline-none"
>
{#each REASONING_EFFORT_LEVELS as level (level.value)}
{#each reasoning.levels as level (level.value)}
{@const tokenLabel = reasoning.tokenLabel(level)}
<button
type="button"
class="flex w-full cursor-pointer items-center gap-3 rounded-md px-2 py-1.75 text-left text-sm transition-colors hover:bg-accent"
class:bg-accent={isSelected(level)}
onclick={() => handleSelection(level)}
class:bg-accent={reasoning.isSelected(level)}
onclick={() => {
reasoning.select(level);
subOpen = false;
}}
>
{#if isSelected(level)}
{#if reasoning.isSelected(level)}
<Check class="h-4 w-4 shrink-0 text-foreground" />
{:else}
<div class="h-4 w-4 shrink-0"></div>
@@ -103,11 +53,9 @@
<span class="flex-1">{level.label}</span>
{#if !level.isOff}
{#if tokenLabel}
<span class="text-[11px] text-muted-foreground opacity-60">
{REASONING_EFFORT_TOKENS[level.value] === -1
? 'Unlimited'
: `Max ${REASONING_EFFORT_TOKENS[level.value].toLocaleString()} tokens`}
{tokenLabel}
</span>
{/if}
tools/ui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActionAdd/ChatFormActionAddSheet.svelte
+72 -5
View File
@@ -10,10 +10,18 @@
import { ATTACHMENT_FILE_ITEMS } from '$lib/constants/attachment-menu';
import { useAttachmentMenu } from '$lib/hooks/use-attachment-menu.svelte';
import { useToolsPanel } from '$lib/hooks/use-tools-panel.svelte';
import { useReasoningMenu } from '$lib/hooks/use-reasoning-menu.svelte';
import { conversationsStore } from '$lib/stores/conversations.svelte';
import { mcpStore } from '$lib/stores/mcp.svelte';
import { McpLogo } from '$lib/components/app';
import { PencilRuler, ChevronDown, ChevronRight } from '@lucide/svelte';
import {
PencilRuler,
ChevronDown,
ChevronRight,
Lightbulb,
LightbulbOff,
Check
} from '@lucide/svelte';
import { HealthCheckStatus } from '$lib/enums';
import { AttachmentAction } from '$lib/enums/attachment.enums';
@@ -48,6 +56,7 @@
}: Props = $props();
let sheetOpen = $state(false);
let reasoningExpanded = $state(false);
let filesExpanded = $state(true);
let toolsExpanded = $state(false);
let mcpExpanded = $state(false);
@@ -67,6 +76,7 @@
);
const toolsPanel = useToolsPanel();
const reasoning = useReasoningMenu();
const sheetItemClass =
'flex w-full items-center gap-3 rounded-md px-3 py-2.5 text-left text-sm transition-colors hover:bg-accent active:bg-accent disabled:cursor-not-allowed disabled:opacity-50';
@@ -74,7 +84,7 @@
const sheetItemRowClass =
'flex w-full items-center justify-between gap-2 rounded-md px-3 py-2 text-left text-sm transition-colors hover:bg-accent';
let visibleMcpServers = $derived(mcpStore.visibleMcpServers);
let mcpServers = $derived(mcpStore.getServers());
</script>
<div class="flex items-center gap-1 {className}">
@@ -91,6 +101,63 @@
</Sheet.Header>
<div class="flex flex-col gap-1 px-1.5 pb-2">
{#if reasoning.modelSupportsThinking}
<Collapsible.Root
open={reasoningExpanded}
onOpenChange={(open) => (reasoningExpanded = open)}
>
<Collapsible.Trigger class={sheetItemClass}>
{#if reasoningExpanded}
<ChevronDown class="h-4 w-4 shrink-0" />
{:else}
<ChevronRight class="h-4 w-4 shrink-0" />
{/if}
{#if reasoning.thinkingEnabled}
<Lightbulb class="h-4 w-4 shrink-0 text-amber-400" />
{:else}
<LightbulbOff class="h-4 w-4 shrink-0 text-muted-foreground" />
{/if}
<span class="flex-1">Reasoning</span>
<span class="text-xs capitalize text-muted-foreground">
{reasoning.thinkingEnabled ? reasoning.currentEffort : 'off'}
</span>
</Collapsible.Trigger>
<Collapsible.Content>
<div class="flex flex-col gap-0.5 pl-4">
{#each reasoning.levels as level (level.value)}
{@const tokenLabel = reasoning.tokenLabel(level)}
<button
type="button"
class={sheetItemRowClass}
class:bg-accent={reasoning.isSelected(level)}
onclick={() => reasoning.select(level)}
>
<div class="flex min-w-0 items-center gap-3">
{#if reasoning.isSelected(level)}
<Check class="h-4 w-4 shrink-0 text-foreground" />
{:else}
<div class="h-4 w-4 shrink-0"></div>
{/if}
<span class="text-sm">{level.label}</span>
</div>
{#if tokenLabel}
<span class="shrink-0 text-[11px] text-muted-foreground opacity-60">
{tokenLabel}
</span>
{/if}
</button>
{/each}
</div>
</Collapsible.Content>
</Collapsible.Root>
{/if}
<Collapsible.Root open={filesExpanded} onOpenChange={(open) => (filesExpanded = open)}>
<Collapsible.Trigger class={sheetItemClass}>
{#if filesExpanded}
@@ -151,13 +218,13 @@
<span class="flex-1">MCP Servers</span>
<span class="text-xs text-muted-foreground">
{visibleMcpServers.length} server{visibleMcpServers.length !== 1 ? 's' : ''}
{mcpServers.length} server{mcpServers.length !== 1 ? 's' : ''}
</span>
</Collapsible.Trigger>
<Collapsible.Content>
<div class="flex flex-col gap-0.5 pl-4">
{#each visibleMcpServers as server (server.id)}
{#each mcpServers as server (server.id)}
{@const healthState = mcpStore.getHealthCheckState(server.id)}
{@const hasError = healthState.status === HealthCheckStatus.ERROR}
{@const displayName = mcpStore.getServerLabel(server)}
@@ -200,7 +267,7 @@
</button>
{/each}
{#if visibleMcpServers.length === 0}
{#if mcpServers.length === 0}
<div class="px-3 py-2 text-center text-sm text-muted-foreground">
No MCP servers configured
</div>
tools/ui/src/lib/components/app/chat/ChatMessages/ChatMessageActions/ChatMessageActionCard/ChatMessageActionCardPermissionRequest.svelte
+1 -1
View File
@@ -21,7 +21,7 @@
<ChatMessageActionCard icon={ShieldQuestion}>
{#snippet message()}
Allow use of <span class="font-semibold">{toolName}</span>{#if serverLabel}
from <span class="font-semibold">{serverLabel}</span>{/if}?
&nbsp;from <span class="font-semibold">{serverLabel}</span>{/if}?
{/snippet}
{#snippet actions()}
tools/ui/src/lib/components/app/dialogs/DialogMcpServerAddNew.svelte
+6 -1
View File
@@ -16,6 +16,7 @@
let newServerUrl = $state('');
let newServerHeaders = $state('');
let newServerUseProxy = $state(false);
let newServerUrlError = $derived.by(() => {
if (!newServerUrl.trim()) return 'URL is required';
try {
@@ -35,6 +36,7 @@
if (!value) {
newServerUrl = '';
newServerHeaders = '';
newServerUseProxy = false;
}
open = value;
onOpenChange?.(value);
@@ -49,7 +51,8 @@
id: newServerId,
enabled: true,
url: newServerUrl.trim(),
headers: newServerHeaders.trim() || undefined
headers: newServerHeaders.trim() || undefined,
useProxy: newServerUseProxy
});
conversationsStore.setMcpServerOverride(newServerId, true);
@@ -74,8 +77,10 @@
<McpServerForm
url={newServerUrl}
headers={newServerHeaders}
useProxy={newServerUseProxy}
onUrlChange={(v) => (newServerUrl = v)}
onHeadersChange={(v) => (newServerHeaders = v)}
onUseProxyChange={(v) => (newServerUseProxy = v)}
urlError={newServerUrl ? newServerUrlError : null}
id="new-server"
/>
tools/ui/src/lib/components/app/mcp/McpServerCard/McpServerCard.svelte
+3 -1
View File
@@ -32,7 +32,9 @@
let isHealthChecking = $derived(healthState.status === HealthCheckStatus.CONNECTING);
let isConnected = $derived(healthState.status === HealthCheckStatus.SUCCESS);
let isError = $derived(healthState.status === HealthCheckStatus.ERROR);
let showSkeleton = $derived(isIdle || isHealthChecking);
// Disabled servers stay IDLE (no startup health check), so the body
// skeleton only applies while a check is running or expected to run.
let showSkeleton = $derived(isHealthChecking || (isIdle && server.enabled));
let errorMessage = $derived(
healthState.status === HealthCheckStatus.ERROR ? healthState.message : undefined
);
tools/ui/src/lib/components/app/settings/SettingsMcpServers.svelte
+11 -4
View File
@@ -22,7 +22,9 @@
let { class: className }: Props = $props();
let servers = $derived(mcpStore.visibleMcpServers);
// Every configured server is listed; `enabled` is an on/off state,
// not a visibility filter, so a disabled server stays toggleable.
let servers = $derived(mcpStore.getServers());
let isAddingServer = $state(false);
@@ -58,9 +60,14 @@
// Each card decides for itself whether to render based on its own
// health-check state, so adding a server only flashes the new card
// (not every other already-loaded card) until its health check resolves.
function isServerPending(serverId: string): boolean {
// Disabled servers never receive a startup health check, so IDLE only
// counts as pending when the server is enabled; otherwise the real card
// renders and keeps the enable toggle reachable.
function isServerPending(serverId: string, enabled: boolean): boolean {
const status = mcpStore.getHealthCheckState(serverId).status;
return status === HealthCheckStatus.IDLE || status === HealthCheckStatus.CONNECTING;
return (
status === HealthCheckStatus.CONNECTING || (status === HealthCheckStatus.IDLE && enabled)
);
}
</script>
@@ -109,7 +116,7 @@
style="grid-template-columns: repeat(auto-fill, minmax(min(32rem, calc(100dvw - 2rem)), 1fr));"
>
{#each servers as server (server.id)}
{#if isServerPending(server.id)}
{#if isServerPending(server.id, server.enabled)}
<McpServerCardSkeleton />
{:else}
<McpServerCard
tools/ui/src/lib/constants/agentic.ts
+1 -2
View File
@@ -6,8 +6,7 @@ export const NEWLINE_SEPARATOR = '\n';
export const DEFAULT_AGENTIC_CONFIG: AgenticConfig = {
enabled: true,
maxTurns: 100,
maxToolPreviewLines: 25
maxTurns: 100
} as const;
export const REASONING_TAGS = {
tools/ui/src/lib/constants/model-id.ts
+3 -1
View File
@@ -24,8 +24,10 @@ export const MODEL_CUSTOM_QUANTIZATION_PREFIX_RE = /^UD$/i;
/**
* Matches a parameter-count segment, e.g. `7B`, `1.5b`, `120M`.
* The optional leading `E` covers effective-parameter sizes, e.g. Gemma's
* `E2B`/`E4B` (MatFormer models sized by resident params).
*/
export const MODEL_PARAMS_RE = /^\d+(\.\d+)?[BbMmKkTt]$/;
export const MODEL_PARAMS_RE = /^[Ee]?\d+(\.\d+)?[BbMmKkTt]$/;
/**
* Matches an activated-parameter-count segment, e.g. `A10B`, `a2.4b`.
tools/ui/src/lib/constants/routes.ts
-1
View File
@@ -8,7 +8,6 @@ export const SETTINGS_SECTION_SLUGS = {
PENALTIES: 'penalties',
AGENTIC: 'agentic',
DEVELOPER: 'developer',
MCP: 'mcp',
TOOLS: 'tools',
IMPORT_EXPORT: 'import-export'
} as const;
tools/ui/src/lib/constants/settings-keys.ts
-1
View File
@@ -59,7 +59,6 @@ export const SETTINGS_KEYS = {
MCP_SERVERS: 'mcpServers',
MCP_REQUEST_TIMEOUT_SECONDS: 'mcpRequestTimeoutSeconds',
AGENTIC_MAX_TURNS: 'agenticMaxTurns',
AGENTIC_MAX_TOOL_PREVIEW_LINES: 'agenticMaxToolPreviewLines',
SHOW_TOOL_CALL_IN_PROGRESS: 'showToolCallInProgress',
// Performance
PRE_ENCODE_CONVERSATION: 'preEncodeConversation',
tools/ui/src/lib/constants/settings-registry.ts
+5 -27
View File
@@ -24,7 +24,6 @@ import type {
SettingsSection
} from '$lib/types';
import { CLI_FLAGS, DEFAULT_MCP_CONFIG } from '$lib/constants';
import McpLogo from '$lib/components/app/mcp/McpLogo.svelte';
import { SETTINGS_KEYS } from './settings-keys';
import { ROUTES, SETTINGS_SECTION_SLUGS } from './routes';
import { TITLE_GENERATION } from './title-generation';
@@ -36,7 +35,6 @@ export const SETTINGS_SECTION_TITLES = {
PENALTIES: 'Penalties',
AGENTIC: 'Agentic',
TOOLS: 'Tools',
MCP: 'MCP',
IMPORT_EXPORT: 'Import/Export',
DEVELOPER: 'Developer'
} as const;
@@ -635,15 +633,15 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
}
},
{
key: SETTINGS_KEYS.AGENTIC_MAX_TOOL_PREVIEW_LINES,
label: 'Max lines per tool preview',
help: 'Number of lines shown in tool output previews (last N lines). Only these previews and the final LLM response persist after the agentic loop completes.',
defaultValue: 25,
key: SETTINGS_KEYS.MCP_REQUEST_TIMEOUT_SECONDS,
label: 'MCP request timeout (seconds)',
help: 'Timeout for individual MCP tool calls.',
defaultValue: DEFAULT_MCP_CONFIG.requestTimeoutSeconds,
type: SettingsFieldType.INPUT,
section: SETTINGS_SECTION_SLUGS.AGENTIC,
isPositiveInteger: true,
sync: {
serverKey: SETTINGS_KEYS.AGENTIC_MAX_TOOL_PREVIEW_LINES,
serverKey: SETTINGS_KEYS.MCP_REQUEST_TIMEOUT_SECONDS,
paramType: SyncableParameterType.NUMBER
}
}
@@ -735,26 +733,6 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
}
}
]
},
[SETTINGS_SECTION_SLUGS.MCP]: {
title: SETTINGS_SECTION_TITLES.MCP,
slug: SETTINGS_SECTION_SLUGS.MCP,
icon: McpLogo,
settings: [
{
key: SETTINGS_KEYS.MCP_REQUEST_TIMEOUT_SECONDS,
label: 'Request timeout (seconds)',
help: 'Default timeout for individual MCP tool calls. Can be overridden per server.',
defaultValue: DEFAULT_MCP_CONFIG.requestTimeoutSeconds,
type: SettingsFieldType.INPUT,
section: SETTINGS_SECTION_SLUGS.MCP,
isPositiveInteger: true,
sync: {
serverKey: SETTINGS_KEYS.MCP_REQUEST_TIMEOUT_SECONDS,
paramType: SyncableParameterType.NUMBER
}
}
]
}
} as const;
tools/ui/src/lib/hooks/use-reasoning-menu.svelte.ts
+96
View File
@@ -0,0 +1,96 @@
import { ReasoningEffort } from '$lib/enums';
import { REASONING_EFFORT_LEVELS } from '$lib/constants/reasoning-effort';
import { REASONING_EFFORT_TOKENS } from '$lib/constants/reasoning-effort-tokens';
import type { ReasoningEffortLevel } from '$lib/types';
import type { DatabaseMessage } from '$lib/types/database';
import {
modelsStore,
checkModelSupportsThinking,
supportsThinking,
propsCacheVersion,
loadedModelIds
} from '$lib/stores/models.svelte';
import { chatStore } from '$lib/stores/chat.svelte';
import { conversationsStore, activeMessages } from '$lib/stores/conversations.svelte';
import { isRouterMode } from '$lib/stores/server.svelte';
export interface UseReasoningMenuReturn {
readonly modelSupportsThinking: boolean;
readonly thinkingEnabled: boolean;
readonly currentEffort: ReasoningEffort;
readonly levels: ReasoningEffortLevel[];
isSelected(level: ReasoningEffortLevel): boolean;
tokenLabel(level: ReasoningEffortLevel): string | null;
select(level: ReasoningEffortLevel): void;
}
/**
* Shared reactive state and helpers for the reasoning effort menu.
*
* Used by both the desktop dropdown (`ChatFormActionAddReasoningSubmenu`)
* and the mobile sheet (`ChatFormActionAddSheet`) to avoid duplicating the
* thinking-support derivation and the effort selection logic.
*/
export function useReasoningMenu(): UseReasoningMenuReturn {
const conversationModel = $derived(
chatStore.getConversationModel(activeMessages() as DatabaseMessage[])
);
// a router chat can carry reasoning from an earlier turn before the props
// cache is primed, so a model that already produced thinking still qualifies
const modelSupportsThinkingFromMessages = $derived.by(() => {
const modelId = isRouterMode() ? modelsStore.selectedModelName || conversationModel : null;
if (!modelId) return false;
return conversationsStore.activeMessages.some(
(m) => m.role === 'assistant' && m.model === modelId && !!m.reasoningContent
);
});
const modelSupportsThinking = $derived.by(() => {
loadedModelIds();
propsCacheVersion();
if (isRouterMode()) {
const modelId = modelsStore.selectedModelName || conversationModel;
return checkModelSupportsThinking(modelId ?? '') || modelSupportsThinkingFromMessages;
}
return supportsThinking() || modelSupportsThinkingFromMessages;
});
const thinkingEnabled = $derived(conversationsStore.getThinkingEnabled());
const currentEffort = $derived(conversationsStore.getReasoningEffort());
return {
get modelSupportsThinking() {
return modelSupportsThinking;
},
get thinkingEnabled() {
return thinkingEnabled;
},
get currentEffort() {
return currentEffort;
},
get levels() {
return REASONING_EFFORT_LEVELS;
},
isSelected(level: ReasoningEffortLevel): boolean {
if (level.isOff) return !thinkingEnabled;
return thinkingEnabled && currentEffort === level.value;
},
tokenLabel(level: ReasoningEffortLevel): string | null {
if (level.isOff) return null;
const tokens = REASONING_EFFORT_TOKENS[level.value];
return tokens === -1 ? 'Unlimited' : `Max ${tokens.toLocaleString()} tokens`;
},
select(level: ReasoningEffortLevel): void {
if (level.isOff) {
conversationsStore.setThinkingEnabled(false);
return;
}
conversationsStore.setThinkingEnabled(true);
conversationsStore.setReasoningEffort(level.value as ReasoningEffort);
}
};
}
tools/ui/src/lib/services/mcp.service.ts
+24 -1
View File
@@ -692,8 +692,31 @@ export class MCPService {
this.createLog(MCPConnectionPhase.INITIALIZING, 'Sending initialize request...')
);
// The SDK timeout only covers the initialize request, not transport.start(),
// which can hang forever on an unreachable host (SSE endpoint wait, WebSocket
// handshake, proxied fetch). This race bounds the whole handshake and closes
// the transport on expiry so the underlying fetch or socket is aborted.
const handshakeTimeoutMs =
serverConfig.handshakeTimeoutMs ?? DEFAULT_MCP_CONFIG.connectionTimeoutMs;
try {
await client.connect(transport);
let handshakeTimer: ReturnType<typeof setTimeout> | undefined;
const handshakeDeadline = new Promise<never>((_, reject) => {
handshakeTimer = setTimeout(() => {
void transport.close().catch(() => {});
reject(new Error(`Connection timed out after ${Math.round(handshakeTimeoutMs / 1000)}s`));
}, handshakeTimeoutMs);
});
try {
await Promise.race([
client.connect(transport, { timeout: handshakeTimeoutMs }),
handshakeDeadline
]);
} finally {
clearTimeout(handshakeTimer);
}
// Transport diagnostics are only for the initial handshake, not long-lived traffic.
stopPhaseLogging();
client.onerror = runtimeErrorHandler;
tools/ui/src/lib/stores/agentic.svelte.ts
+1 -4
View File
@@ -280,16 +280,13 @@ class AgenticStore {
getConfig(settings: SettingsConfigType, perChatOverrides?: McpServerOverride[]): AgenticConfig {
const maxTurns = Number(settings.agenticMaxTurns) || DEFAULT_AGENTIC_CONFIG.maxTurns;
const maxToolPreviewLines =
Number(settings.agenticMaxToolPreviewLines) || DEFAULT_AGENTIC_CONFIG.maxToolPreviewLines;
const hasTools =
mcpStore.hasEnabledServers(perChatOverrides) ||
toolsStore.builtinTools.length > 0 ||
toolsStore.customTools.length > 0;
return {
enabled: hasTools && DEFAULT_AGENTIC_CONFIG.enabled,
maxTurns,
maxToolPreviewLines
maxTurns
};
}
tools/ui/src/lib/stores/chat.svelte.ts
+1 -1
View File
@@ -1334,7 +1334,7 @@ class ChatStore {
}
};
const perChatOverrides = conversationsStore.activeConversation?.mcpServerOverrides;
const perChatOverrides = conversationsStore.getAllMcpServerOverrides();
{
const agenticResult = await agenticStore.runAgenticFlow({
tools/ui/src/lib/stores/conversations.svelte.ts
+23 -30
View File
@@ -243,9 +243,9 @@ class ConversationsStore {
const conversationName = name || `Chat ${new Date().toLocaleString()}`;
const conversation = await DatabaseService.createConversation(conversationName);
// New conversations inherit per-server enabled defaults directly from
// `mcpServers[i].enabled` (see #checkServerEnabled). No per-conversation
// override list needs to be seeded.
// No MCP override list is seeded: getAllMcpServerOverrides resolves
// servers without a per-conversation override to `mcpServers[i].enabled`,
// and only explicit toggles are stored on the conversation.
// Inherit global thinking/reasoning defaults into the new conversation
const thinkingEnabled = this.getThinkingEnabled();
@@ -601,48 +601,41 @@ class ConversationsStore {
*/
/**
/**
* Resolve the per-server enabled value when no active conversation exists.
* The default for new chats is the server's own `enabled` flag in `mcpServers`.
* Resolve the default enabled value for a server: its own `enabled`
* flag in `mcpServers`, so the global on/off state lives in one place.
*/
#getDefaultOverrideForNoConversation(serverId: string): McpServerOverride | undefined {
#getDefaultOverride(serverId: string): McpServerOverride | undefined {
const server = mcpStore.getServers().find((s) => s.id === serverId);
if (!server) return undefined;
return { serverId, enabled: server.enabled };
}
/**
* Default overrides for new chats are derived from `mcpServers[i].enabled`,
* so the global on/off state lives in one place.
*/
#getAllDefaultOverridesForNoConversation(): McpServerOverride[] {
return mcpStore.getServers().map((s) => ({ serverId: s.id, enabled: s.enabled }));
}
/**
* Gets MCP server override for a specific server in the active conversation.
* Falls back to `mcpServers[i].enabled` if no active conversation exists.
* Gets the effective MCP server override for a specific server.
* A per-conversation override wins when present; a server without one
* resolves to its `mcpServers[i].enabled` default.
* @param serverId - The server ID to check
* @returns The override if set, undefined if no matching server
* @returns The effective override, undefined if no matching server
*/
getMcpServerOverride(serverId: string): McpServerOverride | undefined {
if (this.activeConversation) {
return this.activeConversation.mcpServerOverrides?.find(
(o: McpServerOverride) => o.serverId === serverId
);
}
return this.#getDefaultOverrideForNoConversation(serverId);
const override = this.activeConversation?.mcpServerOverrides?.find(
(o: McpServerOverride) => o.serverId === serverId
);
if (override) return override;
return this.#getDefaultOverride(serverId);
}
/**
* Get all MCP server overrides for the current conversation.
* When no active conversation, derives from `mcpServers[i].enabled`.
* Gets the effective override list for the current conversation:
* one entry per configured server, resolved per server. The stored
* per-conversation list is sparse and only holds explicit toggles.
*/
getAllMcpServerOverrides(): McpServerOverride[] {
if (this.activeConversation?.mcpServerOverrides) {
return this.activeConversation.mcpServerOverrides;
}
return this.#getAllDefaultOverridesForNoConversation();
const overrides = this.activeConversation?.mcpServerOverrides;
return mcpStore.getServers().map((s) => {
const override = overrides?.find((o: McpServerOverride) => o.serverId === s.id);
return { serverId: s.id, enabled: override?.enabled ?? s.enabled };
});
}
/**
tools/ui/src/lib/stores/mcp.svelte.ts
+17 -21
View File
@@ -148,15 +148,22 @@ class MCPStore {
enabled: Boolean((entry as { enabled?: unknown })?.enabled),
url,
name: (entry as { name?: string })?.name,
requestTimeoutSeconds:
(entry as { requestTimeoutSeconds?: number })?.requestTimeoutSeconds ??
DEFAULT_MCP_CONFIG.requestTimeoutSeconds,
headers: headers || undefined,
useProxy: Boolean((entry as { useProxy?: unknown })?.useProxy)
} satisfies MCPServerSettingsEntry;
});
}
/**
* Request timeout in milliseconds, read live from the global setting
* so a change in Settings applies to every server immediately.
*/
#requestTimeoutMs(): number {
const seconds =
Number(config().mcpRequestTimeoutSeconds) || DEFAULT_MCP_CONFIG.requestTimeoutSeconds;
return Math.round(seconds * 1000);
}
/**
* Builds server configuration from a settings entry.
*/
@@ -183,7 +190,7 @@ class MCPStore {
url: entry.url,
transport: detectMcpTransportFromUrl(entry.url),
handshakeTimeoutMs: connectionTimeoutMs,
requestTimeoutMs: Math.round(entry.requestTimeoutSeconds * 1000),
requestTimeoutMs: this.#requestTimeoutMs(),
headers,
useProxy: entry.useProxy
};
@@ -191,15 +198,15 @@ class MCPStore {
/**
* Checks if a server is enabled for a given chat.
* Only per-chat overrides (persisted in localStorage for new chats,
* or in IndexedDB for existing conversations) control enabled state.
* A per-chat override wins when present; a server without one resolves
* to its own `enabled` flag in `mcpServers`.
*/
#checkServerEnabled(
server: MCPServerSettingsEntry,
perChatOverrides?: McpServerOverride[]
): boolean {
const override = perChatOverrides?.find((o) => o.serverId === server.id);
return override?.enabled ?? false;
return override?.enabled ?? server.enabled;
}
/**
@@ -230,7 +237,7 @@ class MCPStore {
protocolVersion: DEFAULT_MCP_CONFIG.protocolVersion,
capabilities: DEFAULT_MCP_CONFIG.capabilities,
clientInfo: DEFAULT_MCP_CONFIG.clientInfo,
requestTimeoutMs: Math.round(DEFAULT_MCP_CONFIG.requestTimeoutSeconds * 1000),
requestTimeoutMs: this.#requestTimeoutMs(),
servers
};
}
@@ -500,7 +507,7 @@ class MCPStore {
}
addServer(
serverData: Omit<MCPServerSettingsEntry, 'id' | 'requestTimeoutSeconds'> & { id?: string }
serverData: Omit<MCPServerSettingsEntry, 'id'> & { id?: string }
): MCPServerSettingsEntry {
const servers = this.getServers();
const newServer: MCPServerSettingsEntry = {
@@ -509,8 +516,6 @@ class MCPStore {
url: serverData.url.trim(),
name: serverData.name,
headers: serverData.headers?.trim() || undefined,
requestTimeoutSeconds:
Number(config().mcpRequestTimeoutSeconds) || DEFAULT_MCP_CONFIG.requestTimeoutSeconds,
useProxy: serverData.useProxy
};
settingsStore.updateConfig(SETTINGS_KEYS.MCP_SERVERS, JSON.stringify([...servers, newServer]));
@@ -551,14 +556,6 @@ class MCPStore {
});
}
/**
* MCP servers selectable in chat-add UIs and the settings page,
* in the order they were added to the config.
*/
get visibleMcpServers(): MCPServerSettingsEntry[] {
return this.getServers().filter((server) => server.enabled);
}
async ensureInitialized(perChatOverrides?: McpServerOverride[]): Promise<boolean> {
if (!browser) {
return false;
@@ -1226,7 +1223,6 @@ class MCPStore {
id: string;
enabled: boolean;
url: string;
requestTimeoutSeconds: number;
headers?: string;
}[],
skipIfChecked = true,
@@ -1317,7 +1313,7 @@ class MCPStore {
logs: []
});
const timeoutMs = Math.round(server.requestTimeoutSeconds * 1000);
const timeoutMs = this.#requestTimeoutMs();
const headers = this.parseHeaders(server.headers);
try {
tools/ui/src/lib/stores/tools.svelte.ts
+2 -1
View File
@@ -412,7 +412,8 @@ class ToolsStore {
tools: { name: string; description?: string }[];
}[] {
const result: ReturnType<ToolsStore['getMcpToolsFromHealthChecks']> = [];
for (const server of mcpStore.visibleMcpServers) {
for (const server of mcpStore.getServers()) {
if (!server.enabled) continue;
const health = mcpStore.getHealthCheckState(server.id);
if (health.status === HealthCheckStatus.SUCCESS && health.tools.length > 0) {
result.push({
tools/ui/src/lib/types/agentic.d.ts
Vendored
-1
View File
@@ -15,7 +15,6 @@ import type { DatabaseMessage, DatabaseMessageExtra, McpServerOverride } from '.
export interface AgenticConfig {
enabled: boolean;
maxTurns: number;
maxToolPreviewLines: number;
}
/**
tools/ui/src/lib/types/mcp.d.ts
Vendored
-2
View File
@@ -174,7 +174,6 @@ export interface HealthCheckParams {
id: string;
enabled: boolean;
url: string;
requestTimeoutSeconds: number;
headers?: string;
useProxy?: boolean;
}
@@ -220,7 +219,6 @@ export interface MCPServerDisplayInfo {
export type MCPServerSettingsEntry = MCPServerDisplayInfo & {
enabled: boolean;
requestTimeoutSeconds: number;
headers?: string;
iconUrl?: string;
useProxy?: boolean;
tools/ui/src/lib/utils/mcp.ts
-5
View File
@@ -9,7 +9,6 @@ import {
MimeTypeText
} from '$lib/enums';
import {
DEFAULT_MCP_CONFIG,
MCP_SERVER_ID_PREFIX,
IMAGE_FILE_EXTENSION_REGEX,
CODE_FILE_EXTENSION_REGEX,
@@ -64,7 +63,6 @@ export function detectMcpTransportFromUrl(url: string): MCPTransportType {
/**
* Parses MCP server settings from a JSON string or array.
* Preserves per-server requestTimeoutSeconds if stored, otherwise falls back to the global default.
* @param rawServers - The raw servers to parse
* @returns An empty array if the input is invalid.
*/
@@ -103,9 +101,6 @@ export function parseMcpServerSettings(rawServers: unknown): MCPServerSettingsEn
enabled: Boolean((entry as { enabled?: unknown })?.enabled),
url,
name: (entry as { name?: string })?.name,
requestTimeoutSeconds:
(entry as { requestTimeoutSeconds?: number })?.requestTimeoutSeconds ??
DEFAULT_MCP_CONFIG.requestTimeoutSeconds,
headers: headers || undefined,
useProxy: Boolean((entry as { useProxy?: unknown })?.useProxy)
} satisfies MCPServerSettingsEntry;
tools/ui/tests/unit/model-id-parser.test.ts
+5
View File
@@ -36,6 +36,11 @@ describe('parseModelId', () => {
expect(parseModelId('model-100b:q4_k_m')).toMatchObject({ params: '100B' });
});
it('extracts effective parameters correctly', () => {
expect(parseModelId('model-E4B-BF16')).toMatchObject({ params: 'E4B' });
expect(parseModelId('model-e2b:q4_k_m')).toMatchObject({ params: 'E2B' });
});
it('extracts activated parameters correctly', () => {
expect(parseModelId('model-100B-A10B-BF16')).toMatchObject({ activatedParams: 'A10B' });
expect(parseModelId('model-100B-A10B:Q4_K_M')).toMatchObject({ activatedParams: 'A10B' });
tools/ui/tests/unit/parse-mcp-server-settings.test.ts
+24 -16
View File
@@ -1,6 +1,6 @@
import { describe, expect, it, vi } from 'vitest';
import { parseMcpServerSettings } from '$lib/utils/mcp';
import { DEFAULT_MCP_CONFIG, MCP_SERVER_ID_PREFIX } from '$lib/constants/mcp';
import { MCP_SERVER_ID_PREFIX } from '$lib/constants/mcp';
/**
* Tests for the mcpServers settings parser.
@@ -58,24 +58,16 @@ describe('parseMcpServerSettings', () => {
expect(parsed[2]?.id).toBe('custom-3');
});
it('falls back to the configured default requestTimeoutSeconds only for nullish values', () => {
const fallback = DEFAULT_MCP_CONFIG.requestTimeoutSeconds;
it('does not emit a per-server timeout, the request timeout is a live global setting', () => {
// A stored per-server requestTimeoutSeconds was never editable in
// any UI and froze the global setting at server creation time,
// making the Settings value a no-op for existing servers. The
// parser drops the field so the global applies live everywhere.
const parsed = parseMcpServerSettings(
JSON.stringify([
{ id: 'a', url: 'https://a.test' },
{ id: 'b', url: 'https://b.test', requestTimeoutSeconds: undefined },
{ id: 'c', url: 'https://c.test', requestTimeoutSeconds: 0 },
{ id: 'd', url: 'https://d.test', requestTimeoutSeconds: 45 }
])
JSON.stringify([{ id: 'a', url: 'https://a.test', requestTimeoutSeconds: 45 }])
);
// The parser uses ?? for timeout fallback, which only triggers on
// null/undefined. Explicit 0 is preserved at face value.
expect(parsed[0]?.requestTimeoutSeconds).toBe(fallback);
expect(parsed[1]?.requestTimeoutSeconds).toBe(fallback);
expect(parsed[2]?.requestTimeoutSeconds).toBe(0);
expect(parsed[3]?.requestTimeoutSeconds).toBe(45);
expect(parsed[0]).not.toHaveProperty('requestTimeoutSeconds');
});
it('treats whitespace-only headers strings as undefined', () => {
@@ -108,6 +100,22 @@ describe('parseMcpServerSettings', () => {
expect(parsed[3]?.useProxy).toBe(true);
});
it('keeps disabled entries in the list, enabled is state and never a visibility filter', () => {
// Regression guard for issue #25625: filtering the server list on
// `enabled` hides a toggled-off server from every UI surface with
// no way to re-enable it. Any list derived from this parser must
// contain disabled entries.
const parsed = parseMcpServerSettings(
JSON.stringify([
{ id: 'on', url: 'https://on.test', enabled: true },
{ id: 'off', url: 'https://off.test', enabled: false }
])
);
expect(parsed.map((entry) => entry.id)).toEqual(['on', 'off']);
expect(parsed[1]?.enabled).toBe(false);
});
it('preserves input order when mapping entries', () => {
const source = [
{ id: 'gamma', url: 'https://c.test' },