opencode-rag-plugin 2.13.0 → 2.13.2

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 CHANGED
@@ -310,6 +310,22 @@ Requires Ollama ≥ 0.35. **Disabled by default.** Opt in via `decision.enabled`
310
310
  }
311
311
  ```
312
312
 
313
+ `decision.baseUrl` accepts a root URL, an `/api`-suffixed URL, or an `/v1`-suffixed URL - both suffixes are stripped, so requests always go to `<root>/v1/systemone`. Any server exposing that route works, for example **llama-swap**:
314
+
315
+ ```json
316
+ {
317
+ "decision": {
318
+ "enabled": true,
319
+ "provider": "ollama",
320
+ "baseUrl": "http://127.0.0.1:11437",
321
+ "model": "Clef-Flash",
322
+ "apiKey": "sk-llama-local"
323
+ }
324
+ }
325
+ ```
326
+
327
+ `decision.apiKey` is sent as `Authorization: Bearer <key>` and is required when the server protects its routes (llama-swap's `api_key`).
328
+
313
329
  **Using it well**
314
330
 
315
331
  - Always include a `none` / "cannot be determined" option.
@@ -235,14 +235,21 @@ export interface DecisionConfig {
235
235
  /** Decision provider name. Only "ollama" is supported for now. */
236
236
  provider: string;
237
237
  /**
238
- * Base URL of the Ollama server. Both root URLs (`http://127.0.0.1:11434`)
239
- * and the `/api`-suffixed convention used by the other sections are
240
- * accepted — `/api` is stripped because decision models are served at
241
- * `/v1/systemone`.
238
+ * Base URL of the decision server. Root URLs (`http://127.0.0.1:11434`),
239
+ * the `/api`-suffixed convention used by the other sections, and the
240
+ * `/v1`-suffixed convention used by OpenAI-compatible servers (e.g.
241
+ * llama-swap) are accepted - the suffixes are stripped because decision
242
+ * models are served at `/v1/systemone`.
242
243
  */
243
244
  baseUrl: string;
244
245
  /** Decision model name (`tev1`/`tev1:4b` or `tev1:0.8b`). */
245
246
  model: string;
247
+ /**
248
+ * Optional bearer key for the decision endpoint. Some servers (e.g.
249
+ * llama-swap) protect every route with an API key; when set, requests
250
+ * carry `Authorization: Bearer <apiKey>`.
251
+ */
252
+ apiKey?: string;
246
253
  /** Request timeout in milliseconds. The first call may load the model. */
247
254
  timeoutMs: number;
248
255
  /** Ollama keep_alive value (e.g. "30m", "-1" for keep-in-memory). */
@@ -274,13 +281,13 @@ export interface DecisionConfig {
274
281
  * `destructive` or `malicious` probability reaches `threshold` is escalated to
275
282
  * OpenCode's permission prompt; a clean command executes unchanged.
276
283
  *
277
- * The guard reuses the `decision` backend unless `provider`/`baseUrl`/`model`
278
- * override it here. It is applied through OpenCode V2's permission
284
+ * The guard reuses the `decision` backend unless `provider`/`baseUrl`/`model`/
285
+ * `apiKey` override it here. It is applied through OpenCode V2's permission
279
286
  * `evaluate` hook, so it needs no rule changes in `opencode.json`.
280
287
  */
281
288
  export interface ShellGuardConfig {
282
289
  /**
283
- * Master gate. Default false — the guard is opt-in because every shell
290
+ * Master gate. Default false - the guard is opt-in because every shell
284
291
  * command then waits for the local decision model.
285
292
  */
286
293
  enabled: boolean;
@@ -290,6 +297,8 @@ export interface ShellGuardConfig {
290
297
  baseUrl?: string;
291
298
  /** Model override (defaults to `decision.model`). */
292
299
  model?: string;
300
+ /** API-key override (defaults to `decision.apiKey`). */
301
+ apiKey?: string;
293
302
  /**
294
303
  * Flag a command when `p(destructive)` or `p(malicious)` reaches this
295
304
  * probability (0-1). Default 0.5. Lower = more prompts; tune against the
@@ -651,6 +651,9 @@ export function validateConfig(config) {
651
651
  if (config.decision.maxStateChars != null && config.decision.maxStateChars <= 0) {
652
652
  warnings.push("decision.maxStateChars must be > 0");
653
653
  }
654
+ if (config.decision.apiKey != null && typeof config.decision.apiKey !== "string") {
655
+ warnings.push("decision.apiKey must be a string when set");
656
+ }
654
657
  try {
655
658
  new URL(config.decision.baseUrl);
656
659
  }
@@ -675,7 +678,7 @@ export function validateConfig(config) {
675
678
  if (sg.maxCacheEntries != null && (typeof sg.maxCacheEntries !== "number" || !Number.isFinite(sg.maxCacheEntries) || sg.maxCacheEntries < 0)) {
676
679
  warnings.push("shellGuard.maxCacheEntries must be >= 0");
677
680
  }
678
- for (const key of ["provider", "baseUrl", "model"]) {
681
+ for (const key of ["provider", "baseUrl", "model", "apiKey"]) {
679
682
  if (sg[key] != null && typeof sg[key] !== "string") {
680
683
  warnings.push(`shellGuard.${key} must be a string when set`);
681
684
  }
@@ -42,13 +42,14 @@ export declare class OllamaDecisionProvider implements DecisionProvider {
42
42
  readonly name = "ollama";
43
43
  private readonly baseUrl;
44
44
  private readonly model;
45
+ private readonly apiKey?;
45
46
  private readonly timeoutMs;
46
47
  private readonly keepAlive?;
47
48
  private readonly proxy?;
48
49
  private readonly retryMax;
49
50
  private readonly retryBaseDelayMs;
50
51
  /**
51
- * @param config - Decision-model configuration (endpoint, model, timeout, proxy, retries).
52
+ * @param config - Decision-model configuration (endpoint, model, api key, timeout, proxy, retries).
52
53
  */
53
54
  constructor(config: DecisionConfig);
54
55
  /**
@@ -22,7 +22,21 @@ export const SYSTEMONE_MAX_OPTIONS = 24;
22
22
  */
23
23
  export function resolveSystemOneUrl(baseUrl) {
24
24
  const trimmed = baseUrl.replace(/\/+$/, "");
25
- const root = trimmed.endsWith("/api") ? trimmed.slice(0, -"/api".length) : trimmed;
25
+ // Accept root, the `/api` suffix used by the embedding/description
26
+ // sections, and the `/v1` suffix used by OpenAI-compatible servers
27
+ // (e.g. llama-swap). Both suffixes are stripped because the decision
28
+ // route is `/v1/systemone`.
29
+ let root = trimmed;
30
+ let changed = true;
31
+ while (changed) {
32
+ changed = false;
33
+ for (const suffix of ["/api", "/v1"]) {
34
+ if (root.endsWith(suffix)) {
35
+ root = root.slice(0, -suffix.length).replace(/\/+$/, "");
36
+ changed = true;
37
+ }
38
+ }
39
+ }
26
40
  return `${root}/v1/systemone`;
27
41
  }
28
42
  /** Coerce a number-or-numeric-string to a finite number, else undefined. */
@@ -117,17 +131,19 @@ export class OllamaDecisionProvider {
117
131
  name = "ollama";
118
132
  baseUrl;
119
133
  model;
134
+ apiKey;
120
135
  timeoutMs;
121
136
  keepAlive;
122
137
  proxy;
123
138
  retryMax;
124
139
  retryBaseDelayMs;
125
140
  /**
126
- * @param config - Decision-model configuration (endpoint, model, timeout, proxy, retries).
141
+ * @param config - Decision-model configuration (endpoint, model, api key, timeout, proxy, retries).
127
142
  */
128
143
  constructor(config) {
129
144
  this.baseUrl = config.baseUrl;
130
145
  this.model = config.model;
146
+ this.apiKey = config.apiKey?.trim() || undefined;
131
147
  this.timeoutMs = config.timeoutMs > 0 ? config.timeoutMs : 120000;
132
148
  this.keepAlive = config.keepAlive;
133
149
  this.proxy = config.proxy;
@@ -163,11 +179,14 @@ export class OllamaDecisionProvider {
163
179
  `Shorten \`state\` or ask fewer questions.`);
164
180
  }
165
181
  const url = resolveSystemOneUrl(this.baseUrl);
182
+ const headers = this.apiKey
183
+ ? { Authorization: `Bearer ${this.apiKey}` }
184
+ : {};
166
185
  let lastError;
167
186
  for (let attempt = 0; attempt <= this.retryMax; attempt++) {
168
187
  let response;
169
188
  try {
170
- response = await postJson(url, body, {}, this.timeoutMs, this.proxy, abort);
189
+ response = await postJson(url, body, headers, this.timeoutMs, this.proxy, abort);
171
190
  }
172
191
  catch (err) {
173
192
  lastError = err instanceof Error ? err : new Error(String(err));
@@ -211,6 +230,10 @@ export class OllamaDecisionProvider {
211
230
  */
212
231
  describeHttpError(status, text) {
213
232
  const detail = truncate(text);
233
+ if (status === 401 || status === 403) {
234
+ return (`Decision endpoint unauthorized (${status}) - the server requires an API key. ` +
235
+ `Set \`decision.apiKey\` (e.g. llama-swap's "api_key"). ${detail}`);
236
+ }
214
237
  if (status === 404) {
215
238
  const lower = text.toLowerCase();
216
239
  const modelLower = this.model.toLowerCase();
@@ -223,7 +246,7 @@ export class OllamaDecisionProvider {
223
246
  return (`Decision request not found (404) — tev1 requires Ollama >= 0.35 and the model must be pulled ` +
224
247
  `(\`ollama pull ${this.model}\`). ${detail}`);
225
248
  }
226
- return `Ollama decision request failed (${status}): ${detail}`;
249
+ return `Decision request failed (${status}): ${detail}`;
227
250
  }
228
251
  /** Exponential backoff between retries. */
229
252
  sleep(attempt) {
@@ -197,7 +197,9 @@ export class ShellGuard {
197
197
  };
198
198
  }
199
199
  const { guard, decision, logging } = cfg;
200
- const signature = `${decision.provider}|${decision.baseUrl}|${decision.model}`;
200
+ // The API key participates in the signature so a key change rebuilds the
201
+ // provider (the key never leaves this in-memory cache/state).
202
+ const signature = `${decision.provider}|${decision.baseUrl}|${decision.model}|${decision.apiKey ?? ""}`;
201
203
  const cacheKey = `${signature}\u0000${command}`;
202
204
  const cacheTtlMs = guard.cacheTtlMs ?? 0;
203
205
  if (cacheTtlMs > 0) {
@@ -360,6 +362,7 @@ export function resolveShellGuardRuntimeConfig(cfg, directory) {
360
362
  provider: guard.provider?.trim() || base.provider,
361
363
  baseUrl: guard.baseUrl?.trim() || base.baseUrl,
362
364
  model: guard.model?.trim() || base.model,
365
+ apiKey: guard.apiKey?.trim() || base.apiKey,
363
366
  };
364
367
  const log = resolveLogConfig(cfg);
365
368
  return {
@@ -206,12 +206,13 @@ export const SETTINGS_SECTIONS = [
206
206
  {
207
207
  id: "decision",
208
208
  label: "Decision Model",
209
- description: "Local tev1 model behind the make_decision tool.",
209
+ description: "Local decision model behind the make_decision tool (Ollama /v1/systemone or a compatible server such as llama-swap).",
210
210
  fields: [
211
211
  boolean(["decision", "enabled"], "Enable make_decision"),
212
212
  enumField(["decision", "provider"], "Provider", ["ollama"]),
213
213
  string(["decision", "model"], "Model"),
214
214
  string(["decision", "baseUrl"], "Base URL"),
215
+ string(["decision", "apiKey"], "API key"),
215
216
  number(["decision", "timeoutMs"], "Request timeout (ms)", { minimum: 1 }),
216
217
  string(["decision", "keepAlive"], "Ollama keep-alive", { advanced: true }),
217
218
  boolean(["decision", "routeBeforeAsking"], "Route choices before asking"),
@@ -266,6 +267,7 @@ export const SETTINGS_SECTIONS = [
266
267
  string(["shellGuard", "provider"], "Provider override", { advanced: true }),
267
268
  string(["shellGuard", "baseUrl"], "Base URL override", { advanced: true }),
268
269
  string(["shellGuard", "model"], "Model override", { advanced: true }),
270
+ string(["shellGuard", "apiKey"], "API key override", { advanced: true, scope: "global" }),
269
271
  number(["shellGuard", "cacheTtlMs"], "Verdict cache TTL (ms)", {
270
272
  advanced: true,
271
273
  minimum: 0,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "opencode-rag-plugin",
3
- "version": "2.13.0",
3
+ "version": "2.13.2",
4
4
  "description": "OpenCode plugin for local-first RAG-based semantic code search",
5
5
  "type": "module",
6
6
  "main": "./dist/plugin-entry.js",