Compare commits

..

4 Commits

Author SHA1 Message Date
tim.lingo 718a2e0c06 Merge pull request 'feat(engram): accumulation layer — new nodes to top of stack, not core-identity' (#59) from feat/accumulation-layer into stage
El SDK CI - stage / build-and-test (push) Failing after 8m50s
2026-06-17 18:34:05 +00:00
tim.lingo b6187501fd Merge pull request 'Reconcile live runtime data-integrity fixes onto main (UAF + atomic engram_save)' (#58) from fix/runtime-integrity-reconcile into stage
El SDK CI - stage / build-and-test (push) Failing after 9m32s
2026-06-17 18:33:16 +00:00
Tim Lingo 18e1ab6db1 feat(engram): add accumulation layer (layer 5) — new nodes default to it, not core-identity
El SDK Release / build-and-release (pull_request) Failing after 12m23s
Implements the accumulation layer from the Layered Consciousness architecture
(provisional 64/064,262) and answers the deferred design question. Per the spec
and Will's design: new user-facing nodes (memories, knowledge, conversations) are
created in an accumulation layer at the TOP of the consciousness stack — the engram
the user sees — while the layers below (safety, core-identity, domain, imprint,
suit) shape behavior but are hidden from the user.

- Adds ENGRAM_LAYER_ACCUMULATION (5) + the layer record in engram_init_layers
  (activation_priority 50, suppressible, not injectable, transparent=0).
- engram_node and engram_node_full now assign new nodes to ENGRAM_LAYER_ACCUMULATION.
- ENGRAM_LAYER_DEFAULT stays CORE_IDENTITY ON PURPOSE: it is the fallback for LEGACY
  nodes loaded from snapshots without a layer_id, so existing data (the originator
  corpus) is NEVER migrated. New-nodes-only — the immutable-originator rule.

This is the foundation for fixing the identity-bleed / customer-isolation issue
(user data was landing in Neuron's core-identity layer). The retrieval-side
provenance filter (introspection should compile from accumulation, not the
originator corpus — Persona 64/036,574) is a follow-on, pending the batch-2
Layered Consciousness + Engram spec docs for exact semantics. Compiles clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-17 13:14:57 -05:00
Tim Lingo 2dec76c87a fix(runtime): reconcile live data-integrity fixes onto main (UAF + atomic engram_save)
El SDK Release / build-and-release (pull_request) Failing after 17s
Ports the fixes that until now lived only in the un-versioned el-sdk source the live
macOS soul was hand-built from (captured in the [DO NOT MERGE] live-darwin-runtime
snapshot) FORWARD onto main, faithfully and minimally — without dragging in the
snapshot's deletions of main's newer engram_wm_/engram_load_merge/http_serve_async.

1. UAF (hallucinated/lost-saves root cause): engram_new_id + engram_node_full now use
   el_strdup_persist, NOT el_strdup. el_strdup tracks into the per-request arena that
   el_request_end() frees when the creating HTTP request completes — leaving stored
   nodes with dangling pointers (corrupted ids, 'saved but never listed'). Transplanted
   verbatim from the live runtime; el_strdup_persist sites 19->27, matching live.

2. Atomic engram_save: write <path>.tmp, fflush+fsync, rename() over target (atomic on
   POSIX) so a booting soul's engram_load never reads a truncated/0-byte snapshot — the
   genesis -> nodes=1 -> 63-node-clobber loop. Plus a sparse-write floor: refuse to
   overwrite a >200KB snapshot with one < 1/16 its size. (Validated in isolation:
   harness 11/11; rebuilt+booted the darwin soul, round-tripped 5113 nodes, no clobber.)

The response-truncation fix is already on main (_tl_fs_read_len binary-safe length).
Compiles clean. For Will to build through CI/elb and deploy.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-16 19:46:56 -05:00
3 changed files with 377 additions and 63 deletions
+370 -42
View File
@@ -1475,13 +1475,10 @@ static void http_send_response(int fd, const char* body) {
}
const char* eff_body = is_envelope ? env_body : body;
/* Use max(strlen, fs_read_len). fs_read_len is the real byte count for binary
* files (strlen stops at embedded NULs PNG, WOFF2). strlen is correct AND larger
* when a handler WRAPS fs_read output in a longer text/JSON response (e.g.
* /api/safety-contact returns {"configured":...,"contact": <file>}); using
* fs_read_len alone truncated those responses to the file's length. */
size_t _blen_s = strlen(eff_body);
size_t blen = (_tl_fs_read_len > _blen_s) ? _tl_fs_read_len : _blen_s;
/* Use the real byte count from fs_read if available (handles binary files
* with embedded null bytes PNG, WOFF2, etc.). Fall back to strlen for
* normal text/JSON responses where _tl_fs_read_len is 0. */
size_t blen = (_tl_fs_read_len > 0) ? _tl_fs_read_len : strlen(eff_body);
_tl_fs_read_len = 0; /* consume — one-shot per response */
int head_only = _tl_http_head_only;
@@ -1555,8 +1552,7 @@ static void* http_worker(void* arg) {
/* Copy response out BEFORE arena teardown.
* For binary files, _tl_fs_read_len holds the real byte count
* use memcpy instead of strdup so null bytes are preserved. */
size_t _rlen_s = rs ? strlen(rs) : 0;
size_t rlen = (_tl_fs_read_len > _rlen_s) ? _tl_fs_read_len : _rlen_s;
size_t rlen = _tl_fs_read_len > 0 ? _tl_fs_read_len : (rs ? strlen(rs) : 0);
response = malloc(rlen + 1);
if (response && rs) { memcpy(response, rs, rlen); response[rlen] = '\0'; }
else if (response) { response[0] = '\0'; }
@@ -1803,8 +1799,7 @@ static void* http_worker_v2(void* arg) {
el_val_t hmap = http_build_headers_map(hdr_block ? hdr_block : "");
el_val_t r = h(EL_STR(dispatch_method), EL_STR(path), hmap, EL_STR(body));
const char* rs = EL_CSTR(r);
size_t _rlen_s = rs ? strlen(rs) : 0;
size_t rlen = (_tl_fs_read_len > _rlen_s) ? _tl_fs_read_len : _rlen_s;
size_t rlen = _tl_fs_read_len > 0 ? _tl_fs_read_len : (rs ? strlen(rs) : 0);
response = malloc(rlen + 1);
if (response && rs) { memcpy(response, rs, rlen); response[rlen] = '\0'; }
else if (response) { response[0] = '\0'; }
@@ -1887,6 +1882,83 @@ el_val_t http_serve_v2(el_val_t port, el_val_t handler) {
return 0;
}
/* ── http_serve_async — non-blocking HTTP server ─────────────────────────── */
/* Runs the accept loop in a background pthread, returns immediately so the
* calling EL script can continue (e.g. to run an awareness loop).
*
* El signature: http_serve_async(port, handler) -> Void */
typedef struct { int sock; } HttpServeAsyncArg;
static void* _http_serve_async_loop(void* raw) {
HttpServeAsyncArg* a = (HttpServeAsyncArg*)raw;
int sock = a->sock;
free(a);
while (1) {
struct sockaddr_in6 cli;
socklen_t clen = sizeof(cli);
int cfd = accept(sock, (struct sockaddr*)&cli, &clen);
if (cfd < 0) {
if (errno == EINTR) continue;
perror("accept"); break;
}
pthread_mutex_lock(&_http_conn_mu);
while (_http_conn_active >= HTTP_MAX_CONNS) {
pthread_cond_wait(&_http_conn_cv, &_http_conn_mu);
}
_http_conn_active++;
pthread_mutex_unlock(&_http_conn_mu);
HttpWorkerArg* arg = malloc(sizeof(HttpWorkerArg));
if (!arg) { close(cfd); continue; }
arg->fd = cfd;
pthread_t tid;
if (pthread_create(&tid, NULL, http_worker, arg) != 0) {
close(cfd); free(arg);
pthread_mutex_lock(&_http_conn_mu);
_http_conn_active--;
pthread_cond_signal(&_http_conn_cv);
pthread_mutex_unlock(&_http_conn_mu);
continue;
}
pthread_detach(tid);
}
close(sock);
return NULL;
}
void http_serve_async(el_val_t port, el_val_t handler) {
const char* hname = EL_CSTR(handler);
if (hname && looks_like_string(handler)) {
http_set_handler(handler);
}
int p = (int)port;
if (p <= 0 || p > 65535) { fprintf(stderr, "http_serve_async: invalid port %d\n", p); return; }
int sock = socket(AF_INET6, SOCK_STREAM, 0);
if (sock < 0) { perror("socket"); return; }
int yes = 1; int no = 0;
setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes));
setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, &no, sizeof(no));
struct sockaddr_in6 addr;
memset(&addr, 0, sizeof(addr));
addr.sin6_family = AF_INET6;
addr.sin6_addr = in6addr_any;
addr.sin6_port = htons((uint16_t)p);
if (bind(sock, (struct sockaddr*)&addr, sizeof(addr)) < 0) {
perror("bind"); close(sock); return;
}
if (listen(sock, 64) < 0) { perror("listen"); close(sock); return; }
fprintf(stderr, "[http] async listening on [::]:%d (dual-stack)\n", p);
HttpServeAsyncArg* a = malloc(sizeof(HttpServeAsyncArg));
if (!a) { close(sock); return; }
a->sock = sock;
pthread_t tid;
if (pthread_create(&tid, NULL, _http_serve_async_loop, a) != 0) {
perror("pthread_create"); free(a); close(sock); return;
}
pthread_detach(tid);
/* Returns immediately — caller can now run awareness_run() or any loop. */
}
/* Build the response envelope a 4-arg handler can return. We hand-write
* the JSON so the discriminator key always lands first the runtime's
* http_parse_envelope() detects it via prefix match. headers_json must be
@@ -5959,6 +6031,14 @@ void el_cgi_init(el_val_t name, el_val_t dharma_id, el_val_t principal,
#define ENGRAM_LAYER_DOMAIN 2u
#define ENGRAM_LAYER_IMPRINT 3u
#define ENGRAM_LAYER_SUIT 4u
#define ENGRAM_LAYER_ACCUMULATION 5u
/* New user-facing nodes (memories, knowledge, conversations) are created in the
* accumulation layer the top of the consciousness stack, the engram the user
* sees; every layer below shapes behavior but is hidden from the user (Layered
* Consciousness architecture, app 64/064,262). ENGRAM_LAYER_DEFAULT stays
* core-identity ON PURPOSE: it is the fallback home for LEGACY nodes loaded from
* snapshots without a layer_id, so existing data (the originator corpus) is
* never migrated out of its established layer. New != legacy. */
#define ENGRAM_LAYER_DEFAULT ENGRAM_LAYER_CORE_IDENTITY
/* Pass 3 override floor. Layer 0 nodes that received any background
@@ -6136,6 +6216,20 @@ static void engram_init_layers(EngramStore* g) {
.transparent = 0,
.injectable = 1
};
/* Layer 5 — accumulation. The TOP of the consciousness stack: the default
* home for all new user-facing nodes. This is the engram the user sees;
* every layer below shapes behavior but is hidden from the user. Not
* injectable it is the persistent user accumulation, not a swappable
* overlay. transparent=0: its content is surfaced to introspection (it is
* the user's own knowledge/memory), unlike the lower behavioral layers. */
g->layers[g->layer_count++] = (EngramLayer){
.layer_id = ENGRAM_LAYER_ACCUMULATION,
.name = el_strdup_persist("accumulation"),
.activation_priority = 50,
.suppressible = 1,
.transparent = 0,
.injectable = 0
};
}
static EngramStore* engram_get(void) {
@@ -6329,7 +6423,7 @@ el_val_t engram_node(el_val_t content, el_val_t node_type, el_val_t salience) {
n->last_activated = now;
n->created_at = now;
n->updated_at = now;
n->layer_id = ENGRAM_LAYER_DEFAULT;
n->layer_id = ENGRAM_LAYER_ACCUMULATION; /* new user-facing node → top layer */
g->node_count++;
return el_wrap_str(el_strdup(n->id));
}
@@ -6370,7 +6464,7 @@ el_val_t engram_node_full(el_val_t content, el_val_t node_type, el_val_t label,
n->last_activated = now;
n->created_at = now;
n->updated_at = now;
n->layer_id = ENGRAM_LAYER_DEFAULT;
n->layer_id = ENGRAM_LAYER_ACCUMULATION; /* new user-facing node → top layer */
g->node_count++;
return el_wrap_str(el_strdup(n->id));
}
@@ -7300,48 +7394,28 @@ el_val_t engram_save(el_val_t path) {
jb_putc(&b, '}');
}
jb_puts(&b, "]}");
/* --- Anti-clobber sparse-write floor (NTN engram clobber fix) ---------
* Refuse to overwrite an existing populated snapshot with a drastically
* smaller one. A bad boot that loaded only ~63 identity nodes must never
* be able to clobber a healthy 5000+ node snapshot, regardless of the
* upstream cause (genesis fallback, partial load, etc.). */
{
struct stat _st;
if (stat(p, &_st) == 0 && _st.st_size > 200000 &&
(uint64_t)b.len < (uint64_t)_st.st_size / 16) {
fprintf(stderr,
"[engram_save] REFUSED sparse write: new %zu bytes vs existing "
"%lld bytes (< 1/16) — protecting snapshot %s\n",
fprintf(stderr, "[engram_save] REFUSED sparse write: new %zu vs existing %lld (<1/16) protecting %s\n",
b.len, (long long)_st.st_size, p);
free(b.buf);
return 0;
free(b.buf); return 0;
}
}
/* --- Atomic write: tmp + fsync + rename ------------------------------
* Write to a sibling temp file, fsync it durable, then rename() over the
* target. rename() is atomic on POSIX, so a concurrent reader (a booting
* soul's engram_load) never observes a truncated or 0-byte snapshot
* which was the root of the genesis/clobber loop. */
size_t _plen = strlen(p);
char* _tmp = (char*)malloc(_plen + 5);
if (!_tmp) { free(b.buf); return 0; }
memcpy(_tmp, p, _plen);
memcpy(_tmp + _plen, ".tmp", 5); /* includes NUL */
memcpy(_tmp, p, _plen); memcpy(_tmp + _plen, ".tmp", 5);
FILE* f = fopen(_tmp, "wb");
if (!f) { free(_tmp); free(b.buf); return 0; }
size_t w = fwrite(b.buf, 1, b.len, f);
int wok = (w == b.len);
if (wok) { fflush(f); fsync(fileno(f)); }
fclose(f);
free(b.buf);
fclose(f); free(b.buf);
if (!wok) { unlink(_tmp); free(_tmp); return 0; }
if (rename(_tmp, p) != 0) { unlink(_tmp); free(_tmp); return 0; }
free(_tmp);
return 1;
free(_tmp); return 1;
}
/* Helper: extract a string field from a JSON object substring. */
@@ -7962,6 +8036,257 @@ el_val_t engram_query_range(el_val_t start_ms_v, el_val_t end_ms_v) {
return el_wrap_str(b.buf);
}
/* engram_load_merge — like engram_load but WITHOUT resetting the store.
* Reads a JSON snapshot from `path` and adds any nodes/edges not already
* present in the in-memory graph. Dedup is by node id (for nodes) and by
* (from_id, to_id, relation) tuple (for edges).
*
* Returns (as an EL int) the count of new nodes added. Embeddings are
* intentionally skipped on merged nodes to avoid Ollama delays at runtime;
* auto_link_semantic will handle them when nodes are next activated.
*
* Does not merge layers the in-process layer registry is authoritative. */
el_val_t engram_load_merge(el_val_t path) {
const char* p = EL_CSTR(path);
if (!p || !*p) return 0;
FILE* f = fopen(p, "rb");
if (!f) return 0;
fseek(f, 0, SEEK_END);
long sz = ftell(f);
rewind(f);
if (sz <= 0) { fclose(f); return 0; }
char* data = malloc((size_t)sz + 1);
if (!data) { fclose(f); return 0; }
size_t got = fread(data, 1, (size_t)sz, f);
fclose(f);
data[got] = '\0';
EngramStore* g = engram_get();
int64_t added_nodes = 0;
/* Walk nodes array — skip any node whose id already exists */
const char* nodes_p = json_find_key(data, "nodes");
if (nodes_p) {
nodes_p = eg_skip_ws(nodes_p);
if (*nodes_p == '[') {
nodes_p++;
nodes_p = eg_skip_ws(nodes_p);
while (*nodes_p && *nodes_p != ']') {
if (*nodes_p != '{') { nodes_p++; continue; }
const char* end = json_skip_value(nodes_p);
size_t n = (size_t)(end - nodes_p);
char* obj = malloc(n + 1);
memcpy(obj, nodes_p, n); obj[n] = '\0';
char* nid = eg_get_str_field(obj, "id");
int already = (nid && *nid && engram_find_node(nid) != NULL);
free(nid);
if (!already) {
engram_grow_nodes();
EngramNode* nn = &g->nodes[g->node_count];
memset(nn, 0, sizeof(*nn));
nn->id = eg_get_str_field(obj, "id");
nn->content = eg_get_str_field(obj, "content");
nn->node_type = eg_get_str_field(obj, "node_type");
nn->label = eg_get_str_field(obj, "label");
nn->tier = eg_get_str_field(obj, "tier");
nn->tags = eg_get_str_field(obj, "tags");
nn->metadata = eg_get_str_field(obj, "metadata");
if (!nn->metadata || !*nn->metadata) { free(nn->metadata); nn->metadata = strdup("{}"); }
nn->salience = eg_get_num_field(obj, "salience");
nn->importance = eg_get_num_field(obj, "importance");
nn->confidence = eg_get_num_field(obj, "confidence");
nn->temporal_decay_rate = eg_get_num_field(obj, "temporal_decay_rate");
nn->activation_count = eg_get_int_field(obj, "activation_count");
nn->last_activated = eg_get_int_field(obj, "last_activated");
nn->created_at = eg_get_int_field(obj, "created_at");
nn->updated_at = eg_get_int_field(obj, "updated_at");
nn->background_activation = eg_get_num_field(obj, "background_activation");
nn->working_memory_weight = eg_get_num_field(obj, "working_memory_weight");
if (!isfinite(nn->working_memory_weight) || nn->working_memory_weight < 0.0 || nn->working_memory_weight > 1.0)
nn->working_memory_weight = 0.0; /* clamp corrupt snapshot values */
nn->suppression_count = (int32_t)eg_get_int_field(obj, "suppression_count");
if (json_find_key(obj, "layer_id")) {
nn->layer_id = (uint32_t)eg_get_int_field(obj, "layer_id");
} else {
nn->layer_id = ENGRAM_LAYER_DEFAULT;
}
g->node_count++;
added_nodes++;
}
free(obj);
nodes_p = end;
nodes_p = eg_skip_ws(nodes_p);
if (*nodes_p == ',') { nodes_p++; nodes_p = eg_skip_ws(nodes_p); }
}
}
}
/* Walk edges array — skip if (from_id, to_id, relation) already present */
const char* edges_p = json_find_key(data, "edges");
if (edges_p) {
edges_p = eg_skip_ws(edges_p);
if (*edges_p == '[') {
edges_p++;
edges_p = eg_skip_ws(edges_p);
while (*edges_p && *edges_p != ']') {
if (*edges_p != '{') { edges_p++; continue; }
const char* end = json_skip_value(edges_p);
size_t n = (size_t)(end - edges_p);
char* obj = malloc(n + 1);
memcpy(obj, edges_p, n); obj[n] = '\0';
char* efrom = eg_get_str_field(obj, "from_id");
char* eto = eg_get_str_field(obj, "to_id");
char* erel = eg_get_str_field(obj, "relation");
/* Check for duplicate by scanning existing edges */
int dup = 0;
if (efrom && eto && erel) {
for (int64_t ei = 0; ei < g->edge_count; ei++) {
EngramEdge* ex = &g->edges[ei];
if (ex->from_id && ex->to_id && ex->relation &&
strcmp(ex->from_id, efrom) == 0 &&
strcmp(ex->to_id, eto) == 0 &&
strcmp(ex->relation, erel) == 0) {
dup = 1; break;
}
}
}
if (!dup) {
engram_grow_edges();
EngramEdge* ee = &g->edges[g->edge_count];
memset(ee, 0, sizeof(*ee));
ee->id = eg_get_str_field(obj, "id");
ee->from_id = efrom ? efrom : strdup("");
ee->to_id = eto ? eto : strdup("");
ee->relation = erel ? erel : strdup("");
ee->metadata = eg_get_str_field(obj, "metadata");
if (!ee->metadata || !*ee->metadata) { free(ee->metadata); ee->metadata = strdup("{}"); }
ee->weight = eg_get_num_field(obj, "weight");
ee->confidence = eg_get_num_field(obj, "confidence");
ee->created_at = eg_get_int_field(obj, "created_at");
ee->updated_at = eg_get_int_field(obj, "updated_at");
ee->last_fired = eg_get_int_field(obj, "last_fired");
ee->inhibitory = (int)eg_get_int_field(obj, "inhibitory");
if (json_find_key(obj, "layer_id")) {
ee->layer_id = (uint32_t)eg_get_int_field(obj, "layer_id");
} else {
ee->layer_id = ENGRAM_LAYER_DEFAULT;
}
g->edge_count++;
/* NOTE: efrom/eto/erel ownership transferred to ee above */
efrom = NULL; eto = NULL; erel = NULL;
} else {
free(efrom); free(eto); free(erel);
}
free(obj);
edges_p = end;
edges_p = eg_skip_ws(edges_p);
if (*edges_p == ',') { edges_p++; edges_p = eg_skip_ws(edges_p); }
}
}
}
free(data);
return (el_val_t)added_nodes;
}
el_val_t engram_wm_count(void) {
EngramStore* g = engram_get();
int64_t count = 0;
for (int64_t i = 0; i < g->node_count; i++) {
if (g->nodes[i].working_memory_weight > 0.0) count++;
}
return (el_val_t)count;
}
/* Average working_memory_weight across all promoted nodes (wm > 0).
* Returns the float bit-pattern via el_from_float so EL can use it with
* float_to_str / float_gt. Returns 0.0 when no nodes are promoted.
* Useful in heartbeat ISEs to distinguish "many weak activations" (sparse
* graph, low avg) from "few strong activations" (dense subgraph, high avg).
* Added 2026-06-04 self-review for graph health observability. */
el_val_t engram_wm_avg_weight(void) {
EngramStore* g = engram_get();
double sum = 0.0;
int64_t count = 0;
for (int64_t i = 0; i < g->node_count; i++) {
double w = g->nodes[i].working_memory_weight;
/* Defensive guard: skip any corrupt/out-of-range values so a single
* bad snapshot node doesn't produce a garbage average (e.g. 1.77e+234). */
if (w > 0.0 && w <= 1.0 && isfinite(w)) { sum += w; count++; }
}
double avg = (count > 0) ? (sum / (double)count) : 0.0;
return el_from_float(avg);
}
/* engram_wm_top_json — return top N working-memory nodes (by wm weight) as a
* compact JSON array for ISE heartbeat reporting.
*
* Each element: {"label":"...","node_type":"...","tier":"...","wm":0.42}
*
* Purpose: the heartbeat ISE reports wm_active (count) and wm_avg_weight but
* gives zero visibility into WM *composition* which types/tiers are active.
* After long uptime every WM slot is in steady-state decay+re-promotion so
* wm_promotion ISEs never fire (they only fire on 0>0.1 transitions).
* This function fills the observability gap by snapshotting the current top-N
* WM nodes on every heartbeat. Inserted 2026-06-05 self-review. */
el_val_t engram_wm_top_json(el_val_t n_v) {
int64_t top_n = (int64_t)n_v;
if (top_n <= 0) top_n = 10;
if (top_n > 50) top_n = 50;
EngramStore* g = engram_get();
/* Collect indices of promoted nodes, excluding monitoring noise.
* InternalStateEvent nodes are system-observation artifacts they reflect
* what the daemon is doing, not what it knows. Including them in wm_top
* buries real knowledge (Memory, Knowledge, Belief nodes) under a wall of
* heartbeat/curiosity ISEs, making the heartbeat ISE useless for diagnosing
* WM composition. Filter them out here so wm_top always shows substantive
* content. (2026-06-07 self-review) */
int64_t* idx = malloc((size_t)(g->node_count + 1) * sizeof(int64_t));
if (!idx) return el_wrap_str(el_strdup("[]"));
int64_t mc = 0;
for (int64_t i = 0; i < g->node_count; i++) {
if (g->nodes[i].working_memory_weight > 0.0) {
const char* nt = g->nodes[i].node_type;
if (nt && strcmp(nt, "InternalStateEvent") == 0) continue;
idx[mc++] = i;
}
}
/* Insertion-sort descending by wm weight (mc is typically small). */
for (int64_t i = 1; i < mc; i++) {
int64_t key = idx[i];
double kw = g->nodes[key].working_memory_weight;
int64_t j = i;
while (j > 0 && g->nodes[idx[j-1]].working_memory_weight < kw) {
idx[j] = idx[j-1]; j--;
}
idx[j] = key;
}
int64_t emit = mc < top_n ? mc : top_n;
JsonBuf b; jb_init(&b);
jb_putc(&b, '[');
for (int64_t k = 0; k < emit; k++) {
EngramNode* n = &g->nodes[idx[k]];
if (k > 0) jb_putc(&b, ',');
jb_putc(&b, '{');
jb_puts(&b, "\"label\":");
jb_emit_escaped(&b, n->label ? n->label : "");
jb_puts(&b, ",\"node_type\":");
jb_emit_escaped(&b, n->node_type ? n->node_type : "");
jb_puts(&b, ",\"tier\":");
jb_emit_escaped(&b, n->tier ? n->tier : "");
char tmp[48];
snprintf(tmp, sizeof(tmp), ",\"wm\":%.3f", n->working_memory_weight);
jb_puts(&b, tmp);
jb_putc(&b, '}');
}
free(idx);
jb_putc(&b, ']');
return el_wrap_str(b.buf);
}
#ifdef HAVE_CURL
/* ── DHARMA network ─────────────────────────────────────────────────────────
* Real implementation. Peers are addressed by `dharma_id` either bare
@@ -8602,7 +8927,7 @@ static el_val_t llm_provider_request(const char* url, const char* key,
}
}
static el_val_t llm_chain_call(const char* system_str, const char* user_str) {
static el_val_t llm_chain_call(const char* model_pref, const char* system_str, const char* user_str) {
char url_key[64], key_key[64], fmt_key[64], model_key[64];
for (int i = 0; i < LLM_MAX_PROVIDERS; i++) {
snprintf(url_key, sizeof(url_key), "NEURON_LLM_%d_URL", i);
@@ -8615,6 +8940,7 @@ static el_val_t llm_chain_call(const char* system_str, const char* user_str) {
const char* fmt_s = getenv(fmt_key);
int fmt = (fmt_s && strcmp(fmt_s, "anthropic") == 0) ? 1 : 0;
const char* model = getenv(model_key);
if (!model || !*model) model = model_pref; /* fall back to the caller-requested model */
fprintf(stderr, "[llm] trying provider %d (%s)\n", i, url);
el_val_t result = llm_provider_request(url, key, fmt, model, system_str, user_str);
const char* t = EL_CSTR(result);
@@ -8625,7 +8951,7 @@ static el_val_t llm_chain_call(const char* system_str, const char* user_str) {
const char* api_key = getenv("ANTHROPIC_API_KEY");
if (!api_key || !*api_key) return http_error_json("no LLM providers configured");
fprintf(stderr, "[llm] using legacy ANTHROPIC_API_KEY fallback\n");
return llm_provider_request(LLM_API_URL, api_key, 1, NULL, system_str, user_str);
return llm_provider_request(LLM_API_URL, api_key, 1, model_pref, system_str, user_str);
}
/* Legacy llm_request — kept for backward compat with agentic loop internals */
@@ -8689,14 +9015,16 @@ static el_val_t llm_extract_text(el_val_t resp_val) {
}
el_val_t llm_call(el_val_t model, el_val_t prompt) {
const char* m = EL_CSTR(model);
const char* u = EL_CSTR(prompt); if (!u) u = "";
return llm_chain_call(NULL, u);
return llm_chain_call(m, NULL, u);
}
el_val_t llm_call_system(el_val_t model, el_val_t system_prompt, el_val_t user_prompt) {
const char* m = EL_CSTR(model);
const char* s = EL_CSTR(system_prompt); if (!s) s = "";
const char* u = EL_CSTR(user_prompt); if (!u) u = "";
return llm_chain_call(s, u);
return llm_chain_call(m, s, u);
}
/* ── Tool registry for llm_call_agentic ─────────────────────────────────── */
+7
View File
@@ -176,6 +176,7 @@ el_val_t http_set_handler(el_val_t name);
* existing handlers (e.g. products/web/server.el): it dispatches with
* (method, path, body), hardcodes 200 OK, and auto-detects content type. */
el_val_t http_serve_v2(el_val_t port, el_val_t handler);
void http_serve_async(el_val_t port, el_val_t handler);
el_val_t http_set_handler_v2(el_val_t name);
/* Build an HTTP response envelope. `headers_json` should be a JSON object
@@ -638,6 +639,12 @@ el_val_t engram_list_layers_json(void);
* no nodes promoted to working memory. */
el_val_t engram_compile_layered_json(el_val_t intent, el_val_t depth);
/* ── Working memory ──────────────────────────────────────────────────────────*/
el_val_t engram_wm_count(void);
el_val_t engram_wm_avg_weight(void);
el_val_t engram_wm_top_json(el_val_t n);
el_val_t engram_load_merge(el_val_t path);
/* ── LLM (Anthropic API client) ─────────────────────────────────────────────
* All functions call https://api.anthropic.com/v1/messages with the API key
* from env ANTHROPIC_API_KEY. Default model when empty: claude-sonnet-4-5. */
-21
View File
@@ -1,21 +0,0 @@
#!/bin/sh
# build-soul-darwin.sh — replicate `elb` on macOS/arm64 with clang.
# Proven 2026-06-16: produces a Mach-O arm64 soul that boots and serves :7770.
# The official builder `elb` ships Linux-only (CI); this lets us build + test the
# darwin soul locally (e.g. to validate the atomic engram_save fix in isolation).
#
# Usage: scripts/build-soul-darwin.sh <path-to-neuron/dist> [output-binary]
set -e
DIST="${1:?usage: build-soul-darwin.sh <neuron/dist dir> [out]}"
OUT="${2:-./neuron}"
RT="$(cd "$(dirname "$0")/.." && pwd)/lang/el-compiler/runtime"
B="$(mktemp -d)"
# elc-generated dist modules use C89-style implicit cross-module declarations that
# Apple clang rejects as errors by default; resolve at link, so downgrade them.
CFLAGS="-Wno-implicit-function-declaration -Wno-implicit-int -Wno-int-conversion -I$B -I$DIST -I$RT"
cp "$RT/el_runtime.h" "$B/"
clang -c $CFLAGS "$RT/el_runtime.c" -o "$B/el_runtime.o"
for c in "$DIST"/*.c; do clang -c $CFLAGS "$c" -o "$B/$(basename "$c" .c).o"; done
# NOTE: link *.o once — do not also list el_runtime.o separately (duplicate symbols).
clang "$B"/*.o -o "$OUT" -lcurl -lm
echo "built $OUT"