@openviking/dsh-memory-plugin 0.3.0 → 0.4.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +86 -22
- package/capture.mjs +14 -0
- package/client.mjs +6 -155
- package/config.mjs +36 -166
- package/index.mjs +17 -4
- package/mcp-env.mjs +33 -0
- package/mcp.mjs +2 -33
- package/package.json +9 -6
- package/runtime.mjs +115 -28
- package/servers/mcp-proxy.mjs +13 -19
- package/shared/capture-utils.mjs +146 -7
- package/shared/config-schema.mjs +359 -0
- package/shared/credentials.mjs +238 -102
- package/shared/input-filters.mjs +219 -0
- package/shared/mcp-proxy-config.mjs +171 -9
- package/shared/mcp-proxy-core.mjs +35 -15
- package/shared/ov-http.mjs +112 -0
- package/shared/pending-queue.mjs +41 -5
- package/shared/plugin-config.mjs +239 -29
- package/shared/recall-compress-core.mjs +17 -6
- package/shared/recall-core.mjs +128 -10
- package/shared/uri-guard.mjs +173 -1
- package/shared/workspace-config.mjs +433 -0
- package/shared/workspace-identity.mjs +334 -0
- package/shared/workspace-peer.mjs +189 -8
- package/shared/workspace-registry.mjs +87 -0
- package/skills/openviking-memory/SKILL.md +19 -0
- package/skills.mjs +3 -4
- package/uri-guard.mjs +20 -10
- package/shared/setup-wizard.mjs +0 -114
package/shared/plugin-config.mjs
CHANGED
|
@@ -19,43 +19,52 @@
|
|
|
19
19
|
* Resolution stays env → ovcli.conf plugin.<harness> → ovcli.conf plugin →
|
|
20
20
|
* ov.conf harness section (legacy) → defaults.
|
|
21
21
|
*
|
|
22
|
-
*
|
|
23
|
-
* through `
|
|
24
|
-
*
|
|
25
|
-
*
|
|
26
|
-
*
|
|
27
|
-
* silently does nothing.
|
|
22
|
+
* Every harness resolves through `resolveSettings()` below, and assembles its
|
|
23
|
+
* whole configuration through `buildPluginConfig()`, so one `plugin` section
|
|
24
|
+
* configures all of them and `ov config switch` moves behaviour along with
|
|
25
|
+
* credentials. The knobs themselves are declared once in `config-schema.mjs`;
|
|
26
|
+
* this module decides which files supply them and what is derived from them.
|
|
28
27
|
*/
|
|
29
28
|
|
|
30
|
-
import { readFileSync } from "node:fs";
|
|
31
29
|
import { homedir } from "node:os";
|
|
32
|
-
import { join
|
|
30
|
+
import { join } from "node:path";
|
|
33
31
|
|
|
34
|
-
|
|
32
|
+
import { HARNESS_KEYS, KNOBS, harnessKey, resolveKnobs } from "./config-schema.mjs";
|
|
33
|
+
import {
|
|
34
|
+
buildUserAgent,
|
|
35
|
+
loadCredentialFiles,
|
|
36
|
+
readManifestVersion,
|
|
37
|
+
resolveConnection,
|
|
38
|
+
} from "./credentials.mjs";
|
|
39
|
+
import {
|
|
40
|
+
announcedOverrides,
|
|
41
|
+
loadWorkspaceLayers,
|
|
42
|
+
mergeConfigLayers,
|
|
43
|
+
normalizeWorkspaceConfig,
|
|
44
|
+
projectWorkspaceSettings,
|
|
45
|
+
} from "./workspace-config.mjs";
|
|
46
|
+
import { findWorkspaceRoot, resolveWorkspaceIdentity } from "./workspace-identity.mjs";
|
|
47
|
+
import { resolveEffectivePeerId, resolvePluginPeerId } from "./workspace-peer.mjs";
|
|
48
|
+
import { readEntry } from "./workspace-registry.mjs";
|
|
35
49
|
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
codex: "codex",
|
|
39
|
-
};
|
|
50
|
+
const OVCLI_LAYER = "ovcli.conf";
|
|
51
|
+
const OV_CONF_LAYER = "ov.conf";
|
|
40
52
|
|
|
41
|
-
|
|
42
|
-
try {
|
|
43
|
-
return JSON.parse(readFileSync(path, "utf-8"));
|
|
44
|
-
} catch {
|
|
45
|
-
return null;
|
|
46
|
-
}
|
|
47
|
-
}
|
|
53
|
+
export { HARNESS_KEYS, harnessKey };
|
|
48
54
|
|
|
49
55
|
/**
|
|
50
56
|
* Merge the shared plugin section with the harness-specific override.
|
|
51
57
|
* Returns a flat settings object; unknown keys pass through untouched so a
|
|
52
58
|
* harness can add its own knobs without touching this module.
|
|
59
|
+
*
|
|
60
|
+
* `cliFile` is ovcli.conf already parsed. Callers that have it hand it over so
|
|
61
|
+
* the file is read once per build and, more importantly, so the `plugin`
|
|
62
|
+
* section comes out of the same file the credential chain read — an
|
|
63
|
+
* `OPENVIKING_CONFIG_FILE` that is ovcli-shaped is ovcli.conf, and only
|
|
64
|
+
* `loadCredentialFiles` knows that.
|
|
53
65
|
*/
|
|
54
|
-
export function loadPluginSettings(harness, env = process.env) {
|
|
55
|
-
const
|
|
56
|
-
(env.OPENVIKING_CLI_CONFIG_FILE || DEFAULT_OVCLI_CONF_PATH).replace(/^~/, homedir()),
|
|
57
|
-
);
|
|
58
|
-
const file = tryLoadJson(path);
|
|
66
|
+
export function loadPluginSettings(harness, env = process.env, options = {}) {
|
|
67
|
+
const file = options.cliFile || loadCredentialFiles(env).cliFile;
|
|
59
68
|
const plugin = file && typeof file.plugin === "object" && file.plugin ? file.plugin : {};
|
|
60
69
|
|
|
61
70
|
const shared = {};
|
|
@@ -63,11 +72,212 @@ export function loadPluginSettings(harness, env = process.env) {
|
|
|
63
72
|
if (value && typeof value === "object" && !Array.isArray(value)) continue;
|
|
64
73
|
shared[key] = value;
|
|
65
74
|
}
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
75
|
+
// Both spellings of the harness name find the same override, so copying
|
|
76
|
+
// `claude-code` out of the installer is not a silent no-op.
|
|
77
|
+
const key = harnessKey(harness);
|
|
78
|
+
const scoped = {};
|
|
79
|
+
for (const candidate of [key, key.replace(/_/g, "-")]) {
|
|
80
|
+
const value = candidate && plugin[candidate];
|
|
81
|
+
if (value && typeof value === "object" && !Array.isArray(value)) Object.assign(scoped, value);
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
const settings = { ...shared, ...scoped };
|
|
85
|
+
const cwd = String(options.cwd || "").trim();
|
|
86
|
+
if (!cwd) return settings;
|
|
87
|
+
return { ...settings, ...resolveWorkspaceSettings(cwd, env, options).settings };
|
|
88
|
+
}
|
|
89
|
+
|
|
90
|
+
/**
|
|
91
|
+
* Every knob for one harness, resolved through the whole layer stack.
|
|
92
|
+
*
|
|
93
|
+
* `legacy` is that harness's section in ov.conf, which is where tuning used to
|
|
94
|
+
* live: it stays the lowest configured layer so an existing deployment keeps
|
|
95
|
+
* working, and everything above it comes from files a client may edit.
|
|
96
|
+
*
|
|
97
|
+
* Returns the settings, the names some layer actually supplied (what the
|
|
98
|
+
* `*Configured` flags report), and the raw ovcli.conf merge — a couple of
|
|
99
|
+
* callers still need to know a value came from ovcli.conf rather than ov.conf.
|
|
100
|
+
*/
|
|
101
|
+
export function resolveSettings(harness, options = {}) {
|
|
102
|
+
const { env = process.env, cwd = "", legacy = {}, clientVersion = "", cliFile = null } = options;
|
|
103
|
+
const key = harnessKey(harness);
|
|
104
|
+
const plugin = loadPluginSettings(key, env, { cwd, clientVersion, cliFile });
|
|
105
|
+
const { settings, configured, sources } = resolveKnobs({
|
|
106
|
+
harness: key,
|
|
107
|
+
layers: [
|
|
108
|
+
{ name: OV_CONF_LAYER, data: legacy },
|
|
109
|
+
{ name: OVCLI_LAYER, data: plugin },
|
|
110
|
+
],
|
|
111
|
+
env,
|
|
112
|
+
});
|
|
113
|
+
return { settings, configured, sources, plugin };
|
|
114
|
+
}
|
|
69
115
|
|
|
70
|
-
|
|
116
|
+
/**
|
|
117
|
+
* The workspace layers for one cwd, flattened into harness knobs.
|
|
118
|
+
*
|
|
119
|
+
* Takes the cwd explicitly because a hook's `loadConfig()` runs at module top
|
|
120
|
+
* level, before the payload on stdin has said which directory the session is
|
|
121
|
+
* actually in — the caller resolves this again once it knows.
|
|
122
|
+
*/
|
|
123
|
+
export function resolveWorkspaceSettings(cwd, env = process.env, { clientVersion = "" } = {}) {
|
|
124
|
+
const empty = { settings: {}, root: "", provenance: {}, warnings: [], announced: [] };
|
|
125
|
+
try {
|
|
126
|
+
const { root } = findWorkspaceRoot(cwd, env);
|
|
127
|
+
if (!root) return empty;
|
|
128
|
+
|
|
129
|
+
const { layers, warnings } = loadWorkspaceLayers(root, { clientVersion });
|
|
130
|
+
// The identity is what makes the registry's negative evidence work: without
|
|
131
|
+
// it a directory reused by a different repository inherits the old peer.
|
|
132
|
+
const identity = resolveWorkspaceIdentity({ cwd, env });
|
|
133
|
+
const registry = readEntry(root, { identity, env });
|
|
134
|
+
warnings.push(...registry.warnings);
|
|
135
|
+
if (registry.entry?.settings) layers.push({ layer: "registry", data: registry.entry.settings });
|
|
136
|
+
if (registry.entry?.peer) layers.push({ layer: "registry", data: { peer: registry.entry.peer } });
|
|
137
|
+
if (!layers.length) return { ...empty, root, warnings };
|
|
138
|
+
|
|
139
|
+
const { value, provenance } = mergeConfigLayers(layers, warnings);
|
|
140
|
+
normalizeWorkspaceConfig(value, warnings);
|
|
141
|
+
return {
|
|
142
|
+
settings: projectWorkspaceSettings(value),
|
|
143
|
+
root,
|
|
144
|
+
provenance,
|
|
145
|
+
warnings,
|
|
146
|
+
announced: announcedOverrides(provenance),
|
|
147
|
+
value,
|
|
148
|
+
};
|
|
149
|
+
} catch {
|
|
150
|
+
// A hook must never die over a config file. An unreadable layer is no layer.
|
|
151
|
+
return empty;
|
|
152
|
+
}
|
|
153
|
+
}
|
|
154
|
+
|
|
155
|
+
const SEND_ONLY_WHEN_CONFIGURED = KNOBS.filter((knob) => knob.sendOnlyWhenConfigured);
|
|
156
|
+
|
|
157
|
+
function str(value) {
|
|
158
|
+
return typeof value === "string" && value.trim() ? value.trim() : "";
|
|
159
|
+
}
|
|
160
|
+
|
|
161
|
+
/** The ov.conf block named after this harness, the lowest layer of every chain. */
|
|
162
|
+
function ovConfSection(ovFile, key) {
|
|
163
|
+
const section = key ? ovFile?.[key] : null;
|
|
164
|
+
return section && typeof section === "object" && !Array.isArray(section) ? section : {};
|
|
165
|
+
}
|
|
166
|
+
|
|
167
|
+
/**
|
|
168
|
+
* One harness's whole configuration: the files, the knobs, the peer, and the
|
|
169
|
+
* handful of fields derived from them.
|
|
170
|
+
*
|
|
171
|
+
* Every loader used to write this sequence out by hand and each wrote a
|
|
172
|
+
* slightly different subset, which is how ov.conf's `<harness>.authMode` came
|
|
173
|
+
* to apply on three harnesses only, and why two of the four knobs the server
|
|
174
|
+
* defaults for itself were reported as configured on two harnesses and nowhere
|
|
175
|
+
* else. What is left in a loader after this call is what only that harness
|
|
176
|
+
* knows.
|
|
177
|
+
*
|
|
178
|
+
* The connection — server, key, identity, auth mode — is not a knob: it comes
|
|
179
|
+
* from `resolveConnection`, the same call an MCP proxy's connection comes
|
|
180
|
+
* from, and the `plugin` keys and ov.conf block that name it are read there.
|
|
181
|
+
*
|
|
182
|
+
* `legacy` replaces the ov.conf block this harness would otherwise get for its
|
|
183
|
+
* knobs (dsh's host hands it one of its own); `hostInput` is what an embedding
|
|
184
|
+
* host named for this process, which is the more specific answer and outranks
|
|
185
|
+
* every file; `logFile` is the log basename this harness has always used, and
|
|
186
|
+
* a harness that never had a default log path does not get one invented for
|
|
187
|
+
* it; `rootKeyFallback` is for the harnesses whose chain predates the ovcli.conf
|
|
188
|
+
* pinning and has always ended at `server.root_api_key`.
|
|
189
|
+
*/
|
|
190
|
+
export function buildPluginConfig(harness, {
|
|
191
|
+
cwd = "",
|
|
192
|
+
env = process.env,
|
|
193
|
+
legacy = null,
|
|
194
|
+
manifestUrl = "",
|
|
195
|
+
version = "",
|
|
196
|
+
hostInput = {},
|
|
197
|
+
logFile = "",
|
|
198
|
+
rootKeyFallback = false,
|
|
199
|
+
deriveEffectivePeer = false,
|
|
200
|
+
} = {}) {
|
|
201
|
+
const name = String(harness || "");
|
|
202
|
+
const key = harnessKey(name);
|
|
203
|
+
const workspaceCwd = str(cwd) || process.cwd();
|
|
204
|
+
const files = loadCredentialFiles(env);
|
|
205
|
+
// The version answers two questions: what the User-Agent reports, and whether
|
|
206
|
+
// a workspace file asking for a newer client gets to say so.
|
|
207
|
+
const clientVersion = str(version) || (manifestUrl ? readManifestVersion(manifestUrl) : "");
|
|
208
|
+
const { settings, configured, sources } = resolveSettings(key, {
|
|
209
|
+
env,
|
|
210
|
+
cwd: workspaceCwd,
|
|
211
|
+
legacy: legacy || ovConfSection(files.ovFile, key),
|
|
212
|
+
cliFile: files.cliFile,
|
|
213
|
+
clientVersion,
|
|
214
|
+
});
|
|
215
|
+
const connection = resolveConnection(key, { env, files, hostInput, rootKeyFallback });
|
|
216
|
+
const { baseUrl, account, user } = connection;
|
|
217
|
+
const timeoutMs = settings.timeoutMs;
|
|
218
|
+
const peerId = resolvePluginPeerId({
|
|
219
|
+
settings,
|
|
220
|
+
configured,
|
|
221
|
+
sources,
|
|
222
|
+
credentials: connection,
|
|
223
|
+
hostInput: hostInput?.peerId,
|
|
224
|
+
env,
|
|
225
|
+
});
|
|
226
|
+
|
|
227
|
+
const config = {
|
|
228
|
+
...settings,
|
|
229
|
+
harness: name,
|
|
230
|
+
clientVersion,
|
|
231
|
+
userAgent: buildUserAgent(name, clientVersion),
|
|
232
|
+
|
|
233
|
+
baseUrl,
|
|
234
|
+
mcpUrl: connection.mcpUrl,
|
|
235
|
+
apiKey: connection.apiKey,
|
|
236
|
+
account,
|
|
237
|
+
user,
|
|
238
|
+
authMode: connection.authMode,
|
|
239
|
+
sendIdentityHeaders: connection.sendIdentityHeaders,
|
|
240
|
+
|
|
241
|
+
credentialSource: connection.credentialSource,
|
|
242
|
+
apiKeySource: connection.apiKeySource,
|
|
243
|
+
credentialPath: connection.credentialPath,
|
|
244
|
+
hasApiKey: connection.hasApiKey,
|
|
245
|
+
configPath: connection.cliPath || connection.ovPath || null,
|
|
246
|
+
cliConfigPath: connection.cliPath,
|
|
247
|
+
ovConfigPath: connection.ovPath,
|
|
248
|
+
cliPath: connection.cliPath,
|
|
249
|
+
ovPath: connection.ovPath,
|
|
250
|
+
|
|
251
|
+
peerId,
|
|
252
|
+
explicitPeerId: peerId,
|
|
253
|
+
|
|
254
|
+
// A write gets a longer budget than a read, and a digest has to finish
|
|
255
|
+
// inside the recall request it belongs to, so both fall back to a knob.
|
|
256
|
+
captureTimeoutMs: settings.captureTimeoutMs || Math.max(timeoutMs * 2, 30000),
|
|
257
|
+
recallCompressTimeoutMs: settings.recallCompressTimeoutMs
|
|
258
|
+
|| Math.max(1000, settings.recallTimeoutMs - 10000),
|
|
259
|
+
debugLogPath: settings.debugLogPath
|
|
260
|
+
|| (logFile ? join(homedir(), ".openviking", "logs", logFile) : ""),
|
|
261
|
+
|
|
262
|
+
// The spellings these fields carried before every harness shared a loader.
|
|
263
|
+
endpoint: baseUrl,
|
|
264
|
+
accountId: account,
|
|
265
|
+
userId: user,
|
|
266
|
+
requestTimeoutMs: timeoutMs,
|
|
267
|
+
};
|
|
268
|
+
for (const knob of SEND_ONLY_WHEN_CONFIGURED) {
|
|
269
|
+
config[`${knob.name}Configured`] = configured.has(knob.name);
|
|
270
|
+
}
|
|
271
|
+
|
|
272
|
+
// Only for the harnesses that resolve the peer once at load. The rest derive
|
|
273
|
+
// it per session, against the directory the payload names, and an MCP proxy
|
|
274
|
+
// must not derive one from wherever it happened to be launched.
|
|
275
|
+
if (deriveEffectivePeer) {
|
|
276
|
+
const effectivePeer = resolveEffectivePeerId({ cfg: config, cwd: workspaceCwd, env });
|
|
277
|
+
config.effectivePeer = effectivePeer;
|
|
278
|
+
config.legacyPeerId = effectivePeer.legacyPeerId;
|
|
279
|
+
}
|
|
280
|
+
return config;
|
|
71
281
|
}
|
|
72
282
|
|
|
73
283
|
const REWRITE_MODES = new Set(["off", "client", "server", "auto"]);
|
|
@@ -86,10 +86,16 @@ export function repairDigestUris(digest, validUris = []) {
|
|
|
86
86
|
lines.push(line);
|
|
87
87
|
continue;
|
|
88
88
|
}
|
|
89
|
+
let tokenized = line;
|
|
90
|
+
for (const uri of [...valid].sort((a, b) => b.length - a.length)) {
|
|
91
|
+
if (/\s/.test(uri)) tokenized = tokenized.split(uri).join(encodeURI(uri));
|
|
92
|
+
}
|
|
89
93
|
let dropped = false;
|
|
90
|
-
const repaired =
|
|
91
|
-
|
|
92
|
-
|
|
94
|
+
const repaired = tokenized.replace(/viking:\/\/[^\s<>"')\]]+/g, (uri) => {
|
|
95
|
+
let decoded = uri;
|
|
96
|
+
try { decoded = decodeURI(uri); } catch { /* keep the original candidate */ }
|
|
97
|
+
if (validSet.has(decoded)) return decoded;
|
|
98
|
+
const nearest = nearestUri(decoded, valid);
|
|
93
99
|
if (nearest) return nearest;
|
|
94
100
|
dropped = true;
|
|
95
101
|
return uri;
|
|
@@ -149,6 +155,7 @@ async function writeCache(path, value) {
|
|
|
149
155
|
export async function compressRecallContext({
|
|
150
156
|
query,
|
|
151
157
|
rendered,
|
|
158
|
+
shortContext = rendered,
|
|
152
159
|
entries = [],
|
|
153
160
|
cfg = {},
|
|
154
161
|
runCompressor,
|
|
@@ -158,7 +165,9 @@ export async function compressRecallContext({
|
|
|
158
165
|
const input = String(rendered || "").trim();
|
|
159
166
|
if (!input) return { status: COMPRESS_EMPTY, context: "" };
|
|
160
167
|
const minChars = Math.max(0, Number(cfg.recallCompressMinInputChars ?? 1500));
|
|
161
|
-
if (input.length < minChars)
|
|
168
|
+
if (input.length < minChars) {
|
|
169
|
+
return { status: COMPRESS_OK, context: String(shortContext ?? input).trim() };
|
|
170
|
+
}
|
|
162
171
|
|
|
163
172
|
const maxInputChars = Math.max(1000, Number(cfg.recallCompressMaxInputChars || 18000));
|
|
164
173
|
const maxBullets = Math.max(1, Number(cfg.recallCompressMaxBullets || 6));
|
|
@@ -171,7 +180,8 @@ export async function compressRecallContext({
|
|
|
171
180
|
});
|
|
172
181
|
const cached = await readCache(cachePath);
|
|
173
182
|
if (cached?.key === key && typeof cached.digest === "string") {
|
|
174
|
-
|
|
183
|
+
const digest = normalizeCompressedContext(cached.digest, 4000, maxBullets);
|
|
184
|
+
if (digest) return { status: COMPRESS_OK, context: digest };
|
|
175
185
|
}
|
|
176
186
|
|
|
177
187
|
const prompt = buildRecallCompressionPrompt({
|
|
@@ -185,9 +195,10 @@ export async function compressRecallContext({
|
|
|
185
195
|
if (!normalized) return { status: COMPRESS_EMPTY, context: "" };
|
|
186
196
|
|
|
187
197
|
const validUris = entries.map((entry) => entry?.uri).filter(Boolean);
|
|
188
|
-
const
|
|
198
|
+
const repaired = repairDigestUris(normalized, validUris.length
|
|
189
199
|
? validUris
|
|
190
200
|
: (input.match(/viking:\/\/[^\s<>"']+/g) || []));
|
|
201
|
+
const digest = normalizeCompressedContext(repaired, 4000, maxBullets);
|
|
191
202
|
if (!digest) return { status: COMPRESS_FAILED, context: "" };
|
|
192
203
|
|
|
193
204
|
await writeCache(cachePath, { key, digest, updatedAt: now || 0 });
|
package/shared/recall-core.mjs
CHANGED
|
@@ -31,6 +31,28 @@ const CODING_QUOTA_WEIGHTS = {
|
|
|
31
31
|
|
|
32
32
|
let userSpaceCache = "";
|
|
33
33
|
|
|
34
|
+
/**
|
|
35
|
+
* Is recall on for this config?
|
|
36
|
+
*
|
|
37
|
+
* The switch has four spellings in the wild: a boolean `autoRecall`, opencode's
|
|
38
|
+
* `{ enabled }` object, dsh and pi's `syncTurns`, and the global `enabled`
|
|
39
|
+
* that turns the whole plugin off. Reading it here rather than in each loader
|
|
40
|
+
* is what keeps it from drifting a fifth time — and any spelling that says off
|
|
41
|
+
* wins, so a config that disables recall under an older name still disables it.
|
|
42
|
+
*/
|
|
43
|
+
export function isRecallEnabled(cfg = {}) {
|
|
44
|
+
return isSwitchOn([cfg.enabled, cfg.autoRecall, cfg.recall, cfg.syncRecall]);
|
|
45
|
+
}
|
|
46
|
+
|
|
47
|
+
/** Any recognised spelling that says off wins; everything else means on. */
|
|
48
|
+
function isSwitchOn(values) {
|
|
49
|
+
for (const value of values) {
|
|
50
|
+
if (value === false) return false;
|
|
51
|
+
if (value && typeof value === "object" && !Array.isArray(value) && value.enabled === false) return false;
|
|
52
|
+
}
|
|
53
|
+
return true;
|
|
54
|
+
}
|
|
55
|
+
|
|
34
56
|
export function estimateTokens(text) {
|
|
35
57
|
return text ? Math.ceil(String(text).length / 4) : 0;
|
|
36
58
|
}
|
|
@@ -338,9 +360,8 @@ async function resolveItemContent(fetchJSON, item, cfg, actorPeerId = "") {
|
|
|
338
360
|
}
|
|
339
361
|
|
|
340
362
|
async function buildFallbackInjectionBlock(fetchJSON, items, cfg, actorPeerId = "", log = () => {}) {
|
|
341
|
-
|
|
342
|
-
|
|
343
|
-
let budgetRemaining = Math.max(200, Number(cfg.recallTokenBudget || 2000));
|
|
363
|
+
const budgetTotal = Math.max(200, Number(cfg.recallTokenBudget || 2000));
|
|
364
|
+
let budgetRemaining = budgetTotal;
|
|
344
365
|
const lines = [
|
|
345
366
|
"<openviking-context>",
|
|
346
367
|
"Relevant context from OpenViking. Use the read MCP tool to expand URIs.",
|
|
@@ -373,15 +394,15 @@ async function buildFallbackInjectionBlock(fetchJSON, items, cfg, actorPeerId =
|
|
|
373
394
|
|
|
374
395
|
lines.push("</openviking-context>");
|
|
375
396
|
|
|
376
|
-
const budgetUsed =
|
|
397
|
+
const budgetUsed = budgetTotal - budgetRemaining;
|
|
377
398
|
log("recall_injection_built", {
|
|
378
399
|
contentItems: contentCount,
|
|
379
400
|
hintItems: hintCount,
|
|
380
401
|
budgetUsed,
|
|
381
|
-
budgetTotal
|
|
402
|
+
budgetTotal,
|
|
382
403
|
});
|
|
383
404
|
|
|
384
|
-
return lines.join("\n");
|
|
405
|
+
return { block: lines.join("\n"), contentCount, hintCount, budgetUsed, stage: "ranked" };
|
|
385
406
|
}
|
|
386
407
|
|
|
387
408
|
const LEGACY_CACHE_TTL_MS = 6 * 60 * 60 * 1000;
|
|
@@ -417,6 +438,31 @@ export async function markContextFaceLegacy(path = stateFile("context-face.json"
|
|
|
417
438
|
await writeJsonFile(path, { legacyUntil: now + LEGACY_CACHE_TTL_MS });
|
|
418
439
|
}
|
|
419
440
|
|
|
441
|
+
/**
|
|
442
|
+
* A `peer_scope` the server rejects is remembered the same way, but unlike the
|
|
443
|
+
* context face this one is not a silent capability probe: dropping the field
|
|
444
|
+
* widens recall from the caller's own peer to the whole user root, so doctor
|
|
445
|
+
* reads this file back and warns.
|
|
446
|
+
*/
|
|
447
|
+
export function peerScopeMemoPath() {
|
|
448
|
+
return stateFile("peer-scope.json");
|
|
449
|
+
}
|
|
450
|
+
|
|
451
|
+
export async function readPeerScopeDowngrade(path = peerScopeMemoPath(), now = Date.now()) {
|
|
452
|
+
const cached = await readJsonFile(path);
|
|
453
|
+
if (!cached?.legacyUntil || Number(cached.legacyUntil) <= now) return null;
|
|
454
|
+
return cached;
|
|
455
|
+
}
|
|
456
|
+
|
|
457
|
+
export async function markPeerScopeDowngrade(scope, status, path = peerScopeMemoPath(), now = Date.now()) {
|
|
458
|
+
await writeJsonFile(path, {
|
|
459
|
+
legacyUntil: now + LEGACY_CACHE_TTL_MS,
|
|
460
|
+
scope: String(scope || ""),
|
|
461
|
+
status: Number(status) || 0,
|
|
462
|
+
at: now,
|
|
463
|
+
});
|
|
464
|
+
}
|
|
465
|
+
|
|
420
466
|
function looksLikeUnknownField(res) {
|
|
421
467
|
const text = JSON.stringify(res?.error ?? res?.result ?? res?.detail ?? "").toLowerCase();
|
|
422
468
|
return text.includes("extra") || text.includes("mode") || text.includes("unexpected");
|
|
@@ -558,7 +604,15 @@ async function recallViaEndpoint(fetchJSON, cfg, query, actorPeerId = "", log =
|
|
|
558
604
|
export async function postRecall(fetchJSON, body, opts = {}) {
|
|
559
605
|
const actorPeerId = opts.actorPeerId || "";
|
|
560
606
|
const log = opts.log || (() => {});
|
|
607
|
+
const memoPath = opts.peerScopeMemoPath;
|
|
561
608
|
const request = { ...body };
|
|
609
|
+
|
|
610
|
+
// A remembered downgrade is still a downgrade: recall runs wider than the
|
|
611
|
+
// caller asked for, so the memo doubles as the doctor's evidence.
|
|
612
|
+
if (request.peer_scope && await readPeerScopeDowngrade(memoPath)) {
|
|
613
|
+
delete request.peer_scope;
|
|
614
|
+
}
|
|
615
|
+
|
|
562
616
|
const res = await fetchJSON("/api/v1/search/recall", {
|
|
563
617
|
method: "POST",
|
|
564
618
|
body: JSON.stringify(request),
|
|
@@ -566,9 +620,17 @@ export async function postRecall(fetchJSON, body, opts = {}) {
|
|
|
566
620
|
if (!request.peer_scope || (res.status !== 400 && res.status !== 422)) {
|
|
567
621
|
return res;
|
|
568
622
|
}
|
|
623
|
+
// Only an unknown-field rejection means "this server predates peer_scope".
|
|
624
|
+
// Retrying every other 400/422 without it silently widens the search from
|
|
625
|
+
// the caller's own peer to the whole user root.
|
|
626
|
+
if (!looksLikeUnknownField(res)) {
|
|
627
|
+
log("recall_peer_scope_error", { status: res.status || 0 });
|
|
628
|
+
return res;
|
|
629
|
+
}
|
|
569
630
|
|
|
570
631
|
const downgraded = { ...request };
|
|
571
632
|
delete downgraded.peer_scope;
|
|
633
|
+
await markPeerScopeDowngrade(String(request.peer_scope), res.status || 0, memoPath);
|
|
572
634
|
log("recall_peer_scope_downgrade", { status: res.status || 0 });
|
|
573
635
|
return fetchJSON("/api/v1/search/recall", {
|
|
574
636
|
method: "POST",
|
|
@@ -576,11 +638,56 @@ export async function postRecall(fetchJSON, body, opts = {}) {
|
|
|
576
638
|
}, { actorPeerId });
|
|
577
639
|
}
|
|
578
640
|
|
|
641
|
+
/**
|
|
642
|
+
* Recall, plus whatever is still filed under the peer this workspace used
|
|
643
|
+
* before the identity rule changed.
|
|
644
|
+
*
|
|
645
|
+
* Under the default `peer_scope: "all"` this costs nothing: the server's own
|
|
646
|
+
* sweep of `{user_root}/peers` already reaches the old peer's memories. Under
|
|
647
|
+
* `"actor"` that sweep is off by definition, so the old peer is asked for
|
|
648
|
+
* separately — as itself, which is both cheaper and wider than a bare
|
|
649
|
+
* cross-peer read (that would need the user id, and reaches memories only).
|
|
650
|
+
*
|
|
651
|
+
* `stage` names the path the block came from, which is what a host needs to
|
|
652
|
+
* tell "the server had nothing" (`no_results`) from "the score threshold
|
|
653
|
+
* dropped everything" (`filtered_out`) when the block is empty, and a
|
|
654
|
+
* `server_assembled` block from a `ranked` one when it is not.
|
|
655
|
+
*/
|
|
656
|
+
export async function buildRecallBlockDetailed(fetchJSON, cfg, query, options = {}) {
|
|
657
|
+
const primary = await recallForPeer(fetchJSON, cfg, query, options);
|
|
658
|
+
|
|
659
|
+
const legacyPeerId = String(options.legacyPeerId || "").trim();
|
|
660
|
+
const actorPeerId = options.actorPeerId ?? cfg.peerId ?? "";
|
|
661
|
+
if (cfg.recallPeerScope !== "actor" || !legacyPeerId || legacyPeerId === actorPeerId) return primary;
|
|
662
|
+
|
|
663
|
+
const log = options.log || (() => {});
|
|
664
|
+
const legacy = await recallForPeer(fetchJSON, cfg, query, { ...options, actorPeerId: legacyPeerId });
|
|
665
|
+
if (!legacy.block) return primary;
|
|
666
|
+
log("recall_legacy_peer_hit", { legacyPeerId });
|
|
667
|
+
return {
|
|
668
|
+
block: primary.block ? `${primary.block}\n${legacy.block}` : legacy.block,
|
|
669
|
+
contentCount: primary.contentCount + legacy.contentCount,
|
|
670
|
+
hintCount: primary.hintCount + legacy.hintCount,
|
|
671
|
+
budgetUsed: primary.budgetUsed + legacy.budgetUsed,
|
|
672
|
+
stage: primary.block ? primary.stage : legacy.stage,
|
|
673
|
+
};
|
|
674
|
+
}
|
|
675
|
+
|
|
676
|
+
/** The block alone, or null when nothing was injectable. */
|
|
579
677
|
export async function buildRecallBlock(fetchJSON, cfg, query, options = {}) {
|
|
678
|
+
const { block } = await buildRecallBlockDetailed(fetchJSON, cfg, query, options);
|
|
679
|
+
return block || null;
|
|
680
|
+
}
|
|
681
|
+
|
|
682
|
+
function emptyRecall(stage) {
|
|
683
|
+
return { block: "", contentCount: 0, hintCount: 0, budgetUsed: 0, stage };
|
|
684
|
+
}
|
|
685
|
+
|
|
686
|
+
async function recallForPeer(fetchJSON, cfg, query, options = {}) {
|
|
580
687
|
const actorPeerId = options.actorPeerId ?? cfg.peerId ?? "";
|
|
581
688
|
const log = options.log || (() => {});
|
|
582
689
|
const trimmed = String(query || "").trim();
|
|
583
|
-
if (!trimmed) return
|
|
690
|
+
if (!trimmed) return emptyRecall("no_results");
|
|
584
691
|
|
|
585
692
|
// Assembly happens server-side when the deployment offers the context face;
|
|
586
693
|
// older servers fall through to /recall, then to raw find.
|
|
@@ -589,12 +696,23 @@ export async function buildRecallBlock(fetchJSON, cfg, query, options = {}) {
|
|
|
589
696
|
actorPeerId,
|
|
590
697
|
log,
|
|
591
698
|
});
|
|
592
|
-
if (serverBlock !== null)
|
|
699
|
+
if (serverBlock !== null) {
|
|
700
|
+
if (!serverBlock) return emptyRecall("no_results");
|
|
701
|
+
// The server rendered one budgeted unit, so nothing here degraded to a
|
|
702
|
+
// URI-only hint and the block's own estimate is what it cost.
|
|
703
|
+
return {
|
|
704
|
+
block: serverBlock,
|
|
705
|
+
contentCount: 1,
|
|
706
|
+
hintCount: 0,
|
|
707
|
+
budgetUsed: estimateTokens(serverBlock),
|
|
708
|
+
stage: "server_assembled",
|
|
709
|
+
};
|
|
710
|
+
}
|
|
593
711
|
|
|
594
712
|
const recallLimit = Math.max(1, Number(cfg.recallLimit || DEFAULT_CONTEXT_LIMIT));
|
|
595
713
|
const perSourceLimit = Math.max(recallLimit * 2, 8);
|
|
596
714
|
const raw = await searchAllSources(fetchJSON, trimmed, perSourceLimit, actorPeerId, log);
|
|
597
|
-
if (raw.length === 0) return
|
|
715
|
+
if (raw.length === 0) return emptyRecall("no_results");
|
|
598
716
|
|
|
599
717
|
const profile = buildQueryProfile(trimmed);
|
|
600
718
|
const scoreThreshold = Number.isFinite(Number(cfg.scoreThreshold)) ? Number(cfg.scoreThreshold) : 0.35;
|
|
@@ -608,6 +726,6 @@ export async function buildRecallBlock(fetchJSON, cfg, query, options = {}) {
|
|
|
608
726
|
items: picked.map((it) => ({ type: it._sourceType, uri: it.uri, score: clampScore(it.score) })),
|
|
609
727
|
});
|
|
610
728
|
|
|
611
|
-
if (picked.length === 0) return
|
|
729
|
+
if (picked.length === 0) return emptyRecall("filtered_out");
|
|
612
730
|
return buildFallbackInjectionBlock(fetchJSON, picked, cfg, actorPeerId, log);
|
|
613
731
|
}
|