@skaft/hamr 0.4.1

package/dist/hamr/providers/parsers/xlam.js

@@ -0,0 +1,119 @@
+/**
+ * xLAM tool-call parser.
+ *
+ * Parses xLAM-format tool calls. The xLAM family uses Hermes-compatible
+ * XML-style tags with JSON tool call objects:
+ *
+ *   <tool_call>
+ *   {"name": "get_weather", "arguments": {"location": "SF", "unit": "celsius"}}
+ *   </tool_call>
+ *
+ * Reference: vLLM
+ *   --tool-call-parser xlam
+ *   vllm/entrypoints/openai/tool_parsers/xlam_tool_parser.py
+ *
+ * Note: xLAM also supports a second format using plain function name then args:
+ *   get_weather
+ *   {"location": "SF", "unit": "celsius"}
+ * This parser handles both formats.
+ */
+import { extractDelimitedBlocks, generateCallId, safeJsonParse, sanitizeReasoningTags } from "./utils.js";
+const PARSER_ID = "xlam";
+const DESCRIPTION = 'xLAM format: <tool_call>{"name":"...","arguments":{...}}</tool_call> or bare function+JSON';
+const FAMILIES = ["xLAM"];
+export const xlamParser = {
+    id: PARSER_ID,
+    description: DESCRIPTION,
+    modelFamilies: FAMILIES,
+    parse(content) {
+        const sanitized = sanitizeReasoningTags(content);
+        // Try <tool_call>...</tool_call> delimiters first
+        const delimited = extractDelimitedBlocks(sanitized, "<tool_call>", "</tool_call>");
+        if (delimited.blocks.length > 0) {
+            const calls = [];
+            for (let i = 0; i < delimited.blocks.length; i++) {
+                const block = delimited.blocks[i].trim();
+                if (!block)
+                    continue;
+                // xLAM may use JSON object inside the tag
+                const parsed = safeJsonParse(block);
+                if (parsed.ok && typeof parsed.value === "object" && parsed.value !== null && !Array.isArray(parsed.value)) {
+                    const obj = parsed.value;
+                    const name = obj.name ?? obj.tool_name;
+                    if (typeof name === "string" && name.trim()) {
+                        calls.push(buildCall(name.trim(), obj, block, i));
+                        continue;
+                    }
+                }
+                // xLAM alternate format: function name on first line, JSON args on subsequent lines
+                const lines = block
+                    .split("\n")
+                    .map((l) => l.trim())
+                    .filter(Boolean);
+                if (lines.length >= 1) {
+                    const fnName = lines[0];
+                    if (/^[a-zA-Z_]\w*$/.test(fnName)) {
+                        const argsJson = lines.slice(1).join("\n");
+                        let args = {};
+                        if (argsJson.trim()) {
+                            const parsedArgs = safeJsonParse(argsJson);
+                            if (parsedArgs.ok &&
+                                typeof parsedArgs.value === "object" &&
+                                parsedArgs.value !== null &&
+                                !Array.isArray(parsedArgs.value)) {
+                                args = parsedArgs.value;
+                            }
+                        }
+                        calls.push({
+                            id: generateCallId(undefined, i + 1),
+                            name: fnName,
+                            arguments: args,
+                            rawSource: block,
+                            parserId: PARSER_ID,
+                        });
+                        continue;
+                    }
+                }
+                // Unrecognized format in block — treat as error
+                return {
+                    ok: false,
+                    parserId: PARSER_ID,
+                    calls: [],
+                    content: sanitized,
+                    error: `xLAM block ${i + 1}: unrecognized format`,
+                };
+            }
+            const nonToolContent = [delimited.before, ...delimited.between, delimited.after]
+                .filter(Boolean)
+                .join("\n")
+                .trim();
+            return { ok: true, parserId: PARSER_ID, calls, content: nonToolContent };
+        }
+        return { ok: true, parserId: PARSER_ID, calls: [], content: sanitized };
+    },
+};
+function buildCall(name, obj, raw, index) {
+    let args = {};
+    const rawArgs = obj.arguments ?? obj.parameters ?? obj.input;
+    if (typeof rawArgs === "string") {
+        const parsed = safeJsonParse(rawArgs);
+        if (parsed.ok && typeof parsed.value === "object" && parsed.value !== null && !Array.isArray(parsed.value)) {
+            args = parsed.value;
+        }
+    }
+    else if (typeof rawArgs === "object" && rawArgs !== null && !Array.isArray(rawArgs)) {
+        args = rawArgs;
+    }
+    return {
+        id: generateCallId((obj.id ?? obj.call_id), index + 1),
+        name,
+        arguments: args,
+        rawSource: raw,
+        parserId: PARSER_ID,
+    };
+}
+// ─── Factory ──────────────────────────────────────────────
+export function createXlamParser() {
+    return xlamParser;
+}
+//# sourceMappingURL=xlam.js.map

package/dist/hamr/providers/parsers/xlam.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"xlam.js","sourceRoot":"","sources":["../../../../src/hamr/providers/parsers/xlam.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAGH,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,aAAa,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC;AAE1G,MAAM,SAAS,GAAG,MAAM,CAAC;AACzB,MAAM,WAAW,GAAG,4FAA4F,CAAC;AACjH,MAAM,QAAQ,GAAG,CAAC,MAAM,CAAC,CAAC;AAE1B,MAAM,CAAC,MAAM,UAAU,GAAmB;IACzC,EAAE,EAAE,SAAS;IACb,WAAW,EAAE,WAAW;IACxB,aAAa,EAAE,QAAQ;IAEvB,KAAK,CAAC,OAAe;QACpB,MAAM,SAAS,GAAG,qBAAqB,CAAC,OAAO,CAAC,CAAC;QAEjD,kDAAkD;QAClD,MAAM,SAAS,GAAG,sBAAsB,CAAC,SAAS,EAAE,aAAa,EAAE,cAAc,CAAC,CAAC;QAEnF,IAAI,SAAS,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjC,MAAM,KAAK,GAAqB,EAAE,CAAC;YAEnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAClD,MAAM,KAAK,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;gBACzC,IAAI,CAAC,KAAK;oBAAE,SAAS;gBAErB,0CAA0C;gBAC1C,MAAM,MAAM,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;gBACpC,IAAI,MAAM,CAAC,EAAE,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC;oBAC5G,MAAM,GAAG,GAAG,MAAM,CAAC,KAAgC,CAAC;oBACpD,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,SAAS,CAAC;oBACvC,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC;wBAC7C,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC;wBAClD,SAAS;oBACV,CAAC;gBACF,CAAC;gBAED,oFAAoF;gBACpF,MAAM,KAAK,GAAG,KAAK;qBACjB,KAAK,CAAC,IAAI,CAAC;qBACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;qBACpB,MAAM,CAAC,OAAO,CAAC,CAAC;gBAClB,IAAI,KAAK,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;oBACvB,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;oBACxB,IAAI,gBAAgB,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;wBACnC,MAAM,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;wBAC3C,IAAI,IAAI,GAA4B,EAAE,CAAC;wBACvC,IAAI,QAAQ,CAAC,IAAI,EAAE,EAAE,CAAC;4BACrB,MAAM,UAAU,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;4BAC3C,IACC,UAAU,CAAC,EAAE;gCACb,OAAO,UAAU,CAAC,KAAK,KAAK,QAAQ;gCACpC,UAAU,CAAC,KAAK,KAAK,IAAI;gCACzB,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,EAC/B,CAAC;gCACF,IAAI,GAAG,UAAU,CAAC,KAAgC,CAAC;4BACpD,CAAC;wBACF,CAAC;wBACD,KAAK,CAAC,IAAI,CAAC;4BACV,EAAE,EAAE,cAAc,CAAC,SAAS,EAAE,CAAC,GAAG,CAAC,CAAC;4BACpC,IAAI,EAAE,MAAM;4BACZ,SAAS,EAAE,IAAI;4BACf,SAAS,EAAE,KAAK;4BAChB,QAAQ,EAAE,SAAS;yBACnB,CAAC,CAAC;wBACH,SAAS;oBACV,CAAC;gBACF,CAAC;gBAED,gDAAgD;gBAChD,OAAO;oBACN,EAAE,EAAE,KAAK;oBACT,QAAQ,EAAE,SAAS;oBACnB,KAAK,EAAE,EAAE;oBACT,OAAO,EAAE,SAAS;oBAClB,KAAK,EAAE,cAAc,CAAC,GAAG,CAAC,uBAAuB;iBACjD,CAAC;YACH,CAAC;YAED,MAAM,cAAc,GAAG,CAAC,SAAS,CAAC,MAAM,EAAE,GAAG,SAAS,CAAC,OAAO,EAAE,SAAS,CAAC,KAAK,CAAC;iBAC9E,MAAM,CAAC,OAAO,CAAC;iBACf,IAAI,CAAC,IAAI,CAAC;iBACV,IAAI,EAAE,CAAC;YAET,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,cAAc,EAAE,CAAC;QAC1E,CAAC;QAED,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,KAAK,EAAE,EAAE,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC;IACzE,CAAC;CACD,CAAC;AAEF,SAAS,SAAS,CAAC,IAAY,EAAE,GAA4B,EAAE,GAAW,EAAE,KAAa;IACxF,IAAI,IAAI,GAA4B,EAAE,CAAC;IACvC,MAAM,OAAO,GAAG,GAAG,CAAC,SAAS,IAAI,GAAG,CAAC,UAAU,IAAI,GAAG,CAAC,KAAK,CAAC;IAC7D,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;QACjC,MAAM,MAAM,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC;QACtC,IAAI,MAAM,CAAC,EAAE,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC;YAC5G,IAAI,GAAG,MAAM,CAAC,KAAgC,CAAC;QAChD,CAAC;IACF,CAAC;SAAM,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACvF,IAAI,GAAG,OAAkC,CAAC;IAC3C,CAAC;IACD,OAAO;QACN,EAAE,EAAE,cAAc,CAAC,CAAC,GAAG,CAAC,EAAE,IAAI,GAAG,CAAC,OAAO,CAAuB,EAAE,KAAK,GAAG,CAAC,CAAC;QAC5E,IAAI;QACJ,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,GAAG;QACd,QAAQ,EAAE,SAAS;KACnB,CAAC;AACH,CAAC;AAED,6DAA6D;AAE7D,MAAM,UAAU,gBAAgB;IAC/B,OAAO,UAAU,CAAC;AACnB,CAAC","sourcesContent":["/**\n * xLAM tool-call parser.\n *\n * Parses xLAM-format tool calls. The xLAM family uses Hermes-compatible\n * XML-style tags with JSON tool call objects:\n *\n *   <tool_call>\n *   {\"name\": \"get_weather\", \"arguments\": {\"location\": \"SF\", \"unit\": \"celsius\"}}\n *   </tool_call>\n *\n * Reference: vLLM\n *   --tool-call-parser xlam\n *   vllm/entrypoints/openai/tool_parsers/xlam_tool_parser.py\n *\n * Note: xLAM also supports a second format using plain function name then args:\n *   get_weather\n *   {\"location\": \"SF\", \"unit\": \"celsius\"}\n * This parser handles both formats.\n */\n\nimport type { ParsedToolCall, ToolCallParseResult, ToolCallParser } from \"./types.ts\";\nimport { extractDelimitedBlocks, generateCallId, safeJsonParse, sanitizeReasoningTags } from \"./utils.ts\";\n\nconst PARSER_ID = \"xlam\";\nconst DESCRIPTION = 'xLAM format: <tool_call>{\"name\":\"...\",\"arguments\":{...}}</tool_call> or bare function+JSON';\nconst FAMILIES = [\"xLAM\"];\n\nexport const xlamParser: ToolCallParser = {\n\tid: PARSER_ID,\n\tdescription: DESCRIPTION,\n\tmodelFamilies: FAMILIES,\n\n\tparse(content: string): ToolCallParseResult {\n\t\tconst sanitized = sanitizeReasoningTags(content);\n\n\t\t// Try <tool_call>...</tool_call> delimiters first\n\t\tconst delimited = extractDelimitedBlocks(sanitized, \"<tool_call>\", \"</tool_call>\");\n\n\t\tif (delimited.blocks.length > 0) {\n\t\t\tconst calls: ParsedToolCall[] = [];\n\n\t\t\tfor (let i = 0; i < delimited.blocks.length; i++) {\n\t\t\t\tconst block = delimited.blocks[i].trim();\n\t\t\t\tif (!block) continue;\n\n\t\t\t\t// xLAM may use JSON object inside the tag\n\t\t\t\tconst parsed = safeJsonParse(block);\n\t\t\t\tif (parsed.ok && typeof parsed.value === \"object\" && parsed.value !== null && !Array.isArray(parsed.value)) {\n\t\t\t\t\tconst obj = parsed.value as Record<string, unknown>;\n\t\t\t\t\tconst name = obj.name ?? obj.tool_name;\n\t\t\t\t\tif (typeof name === \"string\" && name.trim()) {\n\t\t\t\t\t\tcalls.push(buildCall(name.trim(), obj, block, i));\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// xLAM alternate format: function name on first line, JSON args on subsequent lines\n\t\t\t\tconst lines = block\n\t\t\t\t\t.split(\"\\n\")\n\t\t\t\t\t.map((l) => l.trim())\n\t\t\t\t\t.filter(Boolean);\n\t\t\t\tif (lines.length >= 1) {\n\t\t\t\t\tconst fnName = lines[0];\n\t\t\t\t\tif (/^[a-zA-Z_]\\w*$/.test(fnName)) {\n\t\t\t\t\t\tconst argsJson = lines.slice(1).join(\"\\n\");\n\t\t\t\t\t\tlet args: Record<string, unknown> = {};\n\t\t\t\t\t\tif (argsJson.trim()) {\n\t\t\t\t\t\t\tconst parsedArgs = safeJsonParse(argsJson);\n\t\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t\tparsedArgs.ok &&\n\t\t\t\t\t\t\t\ttypeof parsedArgs.value === \"object\" &&\n\t\t\t\t\t\t\t\tparsedArgs.value !== null &&\n\t\t\t\t\t\t\t\t!Array.isArray(parsedArgs.value)\n\t\t\t\t\t\t\t) {\n\t\t\t\t\t\t\t\targs = parsedArgs.value as Record<string, unknown>;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\tcalls.push({\n\t\t\t\t\t\t\tid: generateCallId(undefined, i + 1),\n\t\t\t\t\t\t\tname: fnName,\n\t\t\t\t\t\t\targuments: args,\n\t\t\t\t\t\t\trawSource: block,\n\t\t\t\t\t\t\tparserId: PARSER_ID,\n\t\t\t\t\t\t});\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Unrecognized format in block — treat as error\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\tparserId: PARSER_ID,\n\t\t\t\t\tcalls: [],\n\t\t\t\t\tcontent: sanitized,\n\t\t\t\t\terror: `xLAM block ${i + 1}: unrecognized format`,\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tconst nonToolContent = [delimited.before, ...delimited.between, delimited.after]\n\t\t\t\t.filter(Boolean)\n\t\t\t\t.join(\"\\n\")\n\t\t\t\t.trim();\n\n\t\t\treturn { ok: true, parserId: PARSER_ID, calls, content: nonToolContent };\n\t\t}\n\n\t\treturn { ok: true, parserId: PARSER_ID, calls: [], content: sanitized };\n\t},\n};\n\nfunction buildCall(name: string, obj: Record<string, unknown>, raw: string, index: number): ParsedToolCall {\n\tlet args: Record<string, unknown> = {};\n\tconst rawArgs = obj.arguments ?? obj.parameters ?? obj.input;\n\tif (typeof rawArgs === \"string\") {\n\t\tconst parsed = safeJsonParse(rawArgs);\n\t\tif (parsed.ok && typeof parsed.value === \"object\" && parsed.value !== null && !Array.isArray(parsed.value)) {\n\t\t\targs = parsed.value as Record<string, unknown>;\n\t\t}\n\t} else if (typeof rawArgs === \"object\" && rawArgs !== null && !Array.isArray(rawArgs)) {\n\t\targs = rawArgs as Record<string, unknown>;\n\t}\n\treturn {\n\t\tid: generateCallId((obj.id ?? obj.call_id) as string | undefined, index + 1),\n\t\tname,\n\t\targuments: args,\n\t\trawSource: raw,\n\t\tparserId: PARSER_ID,\n\t};\n}\n\n// ─── Factory ──────────────────────────────────────────────\n\nexport function createXlamParser(): ToolCallParser {\n\treturn xlamParser;\n}\n"]}

package/dist/hamr/providers/relay-provider.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Relay model auto-detection.
+ *
+ * The relay is just an OpenAI-compatible HTTP endpoint (llama.cpp / vLLM /
+ * SGLang style). Hamr treats it as a native provider: at startup it queries
+ * the endpoint's `GET /v1/models` and discovers whatever models the relay is
+ * currently serving, so the user never has to hand-list them.
+ *
+ * Ported from synax's `probe-models.ts`: scans the many context-window field
+ * names different OpenAI-compatible servers emit, reads vision capability from
+ * the `capabilities` array (llama.cpp emits `["completion", "multimodal"]`
+ * when `--mmproj` is loaded), and derives a readable display name.
+ *
+ * All failures are swallowed — an unreachable endpoint yields an empty list,
+ * never a crash.
+ */
+/** A model discovered from a relay / OpenAI-compatible endpoint. */
+export interface DiscoveredRelayModel {
+    id: string;
+    displayName: string;
+    contextWindow?: number;
+    maxOutputTokens?: number;
+    supportsThinking: boolean;
+    thinkingLevels: string[];
+    supportsVision?: boolean;
+}
+/**
+ * Fetch available models from an OpenAI-compatible `GET /v1/models` endpoint.
+ *
+ * Returns the discovered models, or an empty array if the endpoint could not
+ * be reached or returned an unrecognized payload. Never throws.
+ */
+export declare function discoverRelayModels(baseUrl: string, apiKey?: string, customHeaders?: Record<string, string>): Promise<DiscoveredRelayModel[]>;
+//# sourceMappingURL=relay-provider.d.ts.map

package/dist/hamr/providers/relay-provider.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"relay-provider.d.ts","sourceRoot":"","sources":["../../../src/hamr/providers/relay-provider.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,oEAAoE;AACpE,MAAM,WAAW,oBAAoB;IACpC,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB;AAgJD;;;;;GAKG;AACH,wBAAsB,mBAAmB,CACxC,OAAO,EAAE,MAAM,EACf,MAAM,CAAC,EAAE,MAAM,EACf,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GACpC,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAyEjC"}

package/dist/hamr/providers/relay-provider.js

@@ -0,0 +1,228 @@
+/**
+ * Relay model auto-detection.
+ *
+ * The relay is just an OpenAI-compatible HTTP endpoint (llama.cpp / vLLM /
+ * SGLang style). Hamr treats it as a native provider: at startup it queries
+ * the endpoint's `GET /v1/models` and discovers whatever models the relay is
+ * currently serving, so the user never has to hand-list them.
+ *
+ * Ported from synax's `probe-models.ts`: scans the many context-window field
+ * names different OpenAI-compatible servers emit, reads vision capability from
+ * the `capabilities` array (llama.cpp emits `["completion", "multimodal"]`
+ * when `--mmproj` is loaded), and derives a readable display name.
+ *
+ * All failures are swallowed — an unreachable endpoint yields an empty list,
+ * never a crash.
+ */
+// Common field names used by various OpenAI-compatible servers for context window.
+// Order roughly by likelihood: llama.cpp forks (max_context_length), vLLM/SGLang
+// (max_model_len), generic (context_length), and others.
+const CONTEXT_WINDOW_FIELDS = [
+    "max_context_length",
+    "max_model_len",
+    "context_length",
+    "max_total_tokens",
+    "total_tokens_capacity",
+    "n_ctx",
+    "max_position_embeddings",
+    "model_max_length",
+    "max_seq_len",
+    "max_sequence_length",
+];
+const KNOWN_THINKING_LEVELS = ["off", "on", "low", "medium", "high", "xhigh"];
+// Discovery is best-effort and must never block startup for long.
+const DISCOVERY_TIMEOUT_MS = 5000;
+function buildHeaders(apiKey, customHeaders) {
+    const headers = {
+        Accept: "application/json",
+        "User-Agent": "hamr/1.0",
+    };
+    if (apiKey) {
+        headers.Authorization = `Bearer ${apiKey}`;
+    }
+    if (customHeaders) {
+        for (const [key, value] of Object.entries(customHeaders)) {
+            headers[key] = value;
+        }
+    }
+    return headers;
+}
+function extractContextWindow(entry) {
+    // Check top-level fields first.
+    for (const field of CONTEXT_WINDOW_FIELDS) {
+        const value = entry[field];
+        if (typeof value === "number" && Number.isFinite(value) && value > 0) {
+            return value;
+        }
+        if (typeof value === "string") {
+            const parsed = Number.parseInt(value, 10);
+            if (!Number.isNaN(parsed) && parsed > 0)
+                return parsed;
+        }
+    }
+    // Some servers (e.g. llama.cpp) nest context-length fields inside meta.
+    const meta = entry.meta;
+    if (meta && typeof meta === "object") {
+        for (const field of CONTEXT_WINDOW_FIELDS) {
+            const value = meta[field];
+            if (typeof value === "number" && Number.isFinite(value) && value > 0) {
+                return value;
+            }
+            if (typeof value === "string") {
+                const parsed = Number.parseInt(value, 10);
+                if (!Number.isNaN(parsed) && parsed > 0)
+                    return parsed;
+            }
+        }
+    }
+    return undefined;
+}
+/**
+ * Derive a human-readable display name from a model ID, stripping common
+ * file extensions and quantization suffixes (gguf / IQ / Q*_K variants).
+ */
+function deriveDisplayName(id) {
+    const cleaned = id
+        .replace(/\.gguf$/i, "")
+        .replace(/[-_]IQ[23456]_\w+$/i, "")
+        .replace(/[-_]Q[234568]_K_\w+$/i, "")
+        .replace(/[-_](f16|f32|q4_0|q4_1|q5_0|q5_1|q8_0)$/i, "");
+    return cleaned.replace(/[-_]/g, " ").replace(/\b\w/g, (c) => c.toUpperCase());
+}
+function stringList(value) {
+    if (!Array.isArray(value))
+        return [];
+    return value.map((item) => String(item).toLowerCase());
+}
+function booleanField(entry, fields) {
+    for (const field of fields) {
+        const value = entry[field];
+        if (typeof value === "boolean")
+            return value;
+        if (typeof value === "string") {
+            const lower = value.toLowerCase();
+            if (["true", "yes", "1"].includes(lower))
+                return true;
+            if (["false", "no", "0"].includes(lower))
+                return false;
+        }
+    }
+    return undefined;
+}
+function extractSupportsVision(entry) {
+    const explicit = booleanField(entry, [
+        "supports_vision",
+        "supportsVision",
+        "vision",
+        "multimodal",
+        "supports_images",
+        "supportsImages",
+    ]);
+    if (explicit !== undefined)
+        return explicit;
+    const meta = entry.meta && typeof entry.meta === "object" ? entry.meta : undefined;
+    if (meta) {
+        const metaExplicit = booleanField(meta, [
+            "supports_vision",
+            "supportsVision",
+            "vision",
+            "multimodal",
+            "supports_images",
+            "supportsImages",
+        ]);
+        if (metaExplicit !== undefined)
+            return metaExplicit;
+    }
+    const haystack = [
+        ...stringList(entry.capabilities),
+        ...stringList(entry.modalities),
+        ...stringList(entry.input_modalities),
+        ...stringList(entry.inputModalities),
+        ...stringList(entry.features),
+        ...(meta ? stringList(meta.capabilities) : []),
+        ...(meta ? stringList(meta.modalities) : []),
+        ...(meta ? stringList(meta.input_modalities) : []),
+        ...(meta ? stringList(meta.inputModalities) : []),
+        ...(meta ? stringList(meta.features) : []),
+    ];
+    if (haystack.some((item) => item === "multimodal" || item === "vision" || item === "image" || item === "images")) {
+        return true;
+    }
+    if (haystack.some((item) => item === "text-only" || item === "text_only")) {
+        return false;
+    }
+    return undefined;
+}
+/**
+ * Fetch available models from an OpenAI-compatible `GET /v1/models` endpoint.
+ *
+ * Returns the discovered models, or an empty array if the endpoint could not
+ * be reached or returned an unrecognized payload. Never throws.
+ */
+export async function discoverRelayModels(baseUrl, apiKey, customHeaders) {
+    const cleanBaseUrl = baseUrl.replace(/\/+$/, "");
+    const headers = buildHeaders(apiKey, customHeaders);
+    const controller = new AbortController();
+    const timeout = setTimeout(() => controller.abort(), DISCOVERY_TIMEOUT_MS);
+    try {
+        const res = await fetch(`${cleanBaseUrl}/models`, {
+            method: "GET",
+            headers,
+            signal: controller.signal,
+        });
+        if (!res.ok)
+            return [];
+        const body = (await res.json());
+        // Standard OpenAI /v1/models response: { object: "list", data: [...] }
+        // Some servers return a bare array instead.
+        const data = body?.data;
+        const entries = Array.isArray(data)
+            ? data
+            : Array.isArray(body)
+                ? body
+                : undefined;
+        if (!entries)
+            return [];
+        const models = [];
+        for (const entry of entries) {
+            const rawId = typeof entry.id === "string" ? entry.id : typeof entry.name === "string" ? entry.name : "";
+            const id = rawId.trim();
+            if (!id)
+                continue;
+            const contextWindow = extractContextWindow(entry);
+            // Prefer server-provided display_name / name, fall back to deriving from ID.
+            const serverName = (typeof entry.display_name === "string" && entry.display_name.trim()) ||
+                (typeof entry.name === "string" && entry.name.trim()) ||
+                "";
+            const displayName = serverName ? serverName : deriveDisplayName(id);
+            const supportsVision = extractSupportsVision(entry);
+            const supportsThinking = typeof entry.supports_thinking === "boolean" ? entry.supports_thinking : false;
+            const apiLevels = Array.isArray(entry.thinking_levels)
+                ? entry.thinking_levels.map((l) => String(l)).filter((l) => KNOWN_THINKING_LEVELS.includes(l))
+                : [];
+            // When the server says thinking is supported but doesn't advertise
+            // specific levels, default to boolean on/off (standard for local models).
+            const thinkingLevels = apiLevels.length > 0 ? apiLevels : supportsThinking ? ["off", "on"] : [];
+            const maxOutputTokens = typeof entry.max_output_tokens === "number" && entry.max_output_tokens > 0
+                ? entry.max_output_tokens
+                : undefined;
+            models.push({
+                id,
+                displayName,
+                contextWindow,
+                maxOutputTokens,
+                supportsThinking,
+                thinkingLevels,
+                supportsVision,
+            });
+        }
+        return models;
+    }
+    catch {
+        return [];
+    }
+    finally {
+        clearTimeout(timeout);
+    }
+}
+//# sourceMappingURL=relay-provider.js.map

package/dist/hamr/providers/relay-provider.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"relay-provider.js","sourceRoot":"","sources":["../../../src/hamr/providers/relay-provider.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAaH,mFAAmF;AACnF,iFAAiF;AACjF,yDAAyD;AACzD,MAAM,qBAAqB,GAAG;IAC7B,oBAAoB;IACpB,eAAe;IACf,gBAAgB;IAChB,kBAAkB;IAClB,uBAAuB;IACvB,OAAO;IACP,yBAAyB;IACzB,kBAAkB;IAClB,aAAa;IACb,qBAAqB;CACrB,CAAC;AAEF,MAAM,qBAAqB,GAAG,CAAC,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;AAE9E,kEAAkE;AAClE,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAElC,SAAS,YAAY,CAAC,MAAe,EAAE,aAAsC;IAC5E,MAAM,OAAO,GAA2B;QACvC,MAAM,EAAE,kBAAkB;QAC1B,YAAY,EAAE,UAAU;KACxB,CAAC;IACF,IAAI,MAAM,EAAE,CAAC;QACZ,OAAO,CAAC,aAAa,GAAG,UAAU,MAAM,EAAE,CAAC;IAC5C,CAAC;IACD,IAAI,aAAa,EAAE,CAAC;QACnB,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,aAAa,CAAC,EAAE,CAAC;YAC1D,OAAO,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC;QACtB,CAAC;IACF,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,oBAAoB,CAAC,KAA8B;IAC3D,gCAAgC;IAChC,KAAK,MAAM,KAAK,IAAI,qBAAqB,EAAE,CAAC;QAC3C,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;QAC3B,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;YACtE,OAAO,KAAK,CAAC;QACd,CAAC;QACD,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC/B,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YAC1C,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC;gBAAE,OAAO,MAAM,CAAC;QACxD,CAAC;IACF,CAAC;IACD,wEAAwE;IACxE,MAAM,IAAI,GAAG,KAAK,CAAC,IAA2C,CAAC;IAC/D,IAAI,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtC,KAAK,MAAM,KAAK,IAAI,qBAAqB,EAAE,CAAC;YAC3C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;YAC1B,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;gBACtE,OAAO,KAAK,CAAC;YACd,CAAC;YACD,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;gBAC/B,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;gBAC1C,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC;oBAAE,OAAO,MAAM,CAAC;YACxD,CAAC;QACF,CAAC;IACF,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED;;;GAGG;AACH,SAAS,iBAAiB,CAAC,EAAU;IACpC,MAAM,OAAO,GAAG,EAAE;SAChB,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,OAAO,CAAC,qBAAqB,EAAE,EAAE,CAAC;SAClC,OAAO,CAAC,uBAAuB,EAAE,EAAE,CAAC;SACpC,OAAO,CAAC,0CAA0C,EAAE,EAAE,CAAC,CAAC;IAC1D,OAAO,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;AAC/E,CAAC;AAED,SAAS,UAAU,CAAC,KAAc;IACjC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACrC,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;AACxD,CAAC;AAED,SAAS,YAAY,CAAC,KAA8B,EAAE,MAAgB;IACrE,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC5B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;QAC3B,IAAI,OAAO,KAAK,KAAK,SAAS;YAAE,OAAO,KAAK,CAAC;QAC7C,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC/B,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;YAClC,IAAI,CAAC,MAAM,EAAE,KAAK,EAAE,GAAG,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC;gBAAE,OAAO,IAAI,CAAC;YACtD,IAAI,CAAC,OAAO,EAAE,IAAI,EAAE,GAAG,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC;gBAAE,OAAO,KAAK,CAAC;QACxD,CAAC;IACF,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED,SAAS,qBAAqB,CAAC,KAA8B;IAC5D,MAAM,QAAQ,GAAG,YAAY,CAAC,KAAK,EAAE;QACpC,iBAAiB;QACjB,gBAAgB;QAChB,QAAQ;QACR,YAAY;QACZ,iBAAiB;QACjB,gBAAgB;KAChB,CAAC,CAAC;IACH,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,QAAQ,CAAC;IAE5C,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,IAAI,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAE,KAAK,CAAC,IAAgC,CAAC,CAAC,CAAC,SAAS,CAAC;IAChH,IAAI,IAAI,EAAE,CAAC;QACV,MAAM,YAAY,GAAG,YAAY,CAAC,IAAI,EAAE;YACvC,iBAAiB;YACjB,gBAAgB;YAChB,QAAQ;YACR,YAAY;YACZ,iBAAiB;YACjB,gBAAgB;SAChB,CAAC,CAAC;QACH,IAAI,YAAY,KAAK,SAAS;YAAE,OAAO,YAAY,CAAC;IACrD,CAAC;IAED,MAAM,QAAQ,GAAG;QAChB,GAAG,UAAU,CAAC,KAAK,CAAC,YAAY,CAAC;QACjC,GAAG,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC;QAC/B,GAAG,UAAU,CAAC,KAAK,CAAC,gBAAgB,CAAC;QACrC,GAAG,UAAU,CAAC,KAAK,CAAC,eAAe,CAAC;QACpC,GAAG,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC;QAC7B,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9C,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5C,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAClD,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QACjD,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;KAC1C,CAAC;IACF,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,YAAY,IAAI,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,OAAO,IAAI,IAAI,KAAK,QAAQ,CAAC,EAAE,CAAC;QAClH,OAAO,IAAI,CAAC;IACb,CAAC;IACD,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,WAAW,IAAI,IAAI,KAAK,WAAW,CAAC,EAAE,CAAC;QAC3E,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACxC,OAAe,EACf,MAAe,EACf,aAAsC;IAEtC,MAAM,YAAY,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACjD,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC;IACpD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,oBAAoB,CAAC,CAAC;IAE3E,IAAI,CAAC;QACJ,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,YAAY,SAAS,EAAE;YACjD,MAAM,EAAE,KAAK;YACb,OAAO;YACP,MAAM,EAAE,UAAU,CAAC,MAAM;SACzB,CAAC,CAAC;QACH,IAAI,CAAC,GAAG,CAAC,EAAE;YAAE,OAAO,EAAE,CAAC;QAEvB,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAY,CAAC;QAE3C,uEAAuE;QACvE,4CAA4C;QAC5C,MAAM,IAAI,GAAI,IAA2B,EAAE,IAAI,CAAC;QAChD,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC;YAClC,CAAC,CAAE,IAAuC;YAC1C,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC;gBACpB,CAAC,CAAE,IAAuC;gBAC1C,CAAC,CAAC,SAAS,CAAC;QACd,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QAExB,MAAM,MAAM,GAA2B,EAAE,CAAC;QAC1C,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC7B,MAAM,KAAK,GAAG,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;YACzG,MAAM,EAAE,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;YACxB,IAAI,CAAC,EAAE;gBAAE,SAAS;YAElB,MAAM,aAAa,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;YAElD,6EAA6E;YAC7E,MAAM,UAAU,GACf,CAAC,OAAO,KAAK,CAAC,YAAY,KAAK,QAAQ,IAAI,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,CAAC;gBACrE,CAAC,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;gBACrD,EAAE,CAAC;YACJ,MAAM,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,iBAAiB,CAAC,EAAE,CAAC,CAAC;YAEpE,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,CAAC,CAAC;YAEpD,MAAM,gBAAgB,GAAG,OAAO,KAAK,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,KAAK,CAAC;YACxG,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,eAAe,CAAC;gBACrD,CAAC,CAAE,KAAK,CAAC,eAA6B,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,qBAAqB,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC;gBAC7G,CAAC,CAAC,EAAE,CAAC;YACN,mEAAmE;YACnE,0EAA0E;YAC1E,MAAM,cAAc,GAAG,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAEhG,MAAM,eAAe,GACpB,OAAO,KAAK,CAAC,iBAAiB,KAAK,QAAQ,IAAI,KAAK,CAAC,iBAAiB,GAAG,CAAC;gBACzE,CAAC,CAAC,KAAK,CAAC,iBAAiB;gBACzB,CAAC,CAAC,SAAS,CAAC;YAEd,MAAM,CAAC,IAAI,CAAC;gBACX,EAAE;gBACF,WAAW;gBACX,aAAa;gBACb,eAAe;gBACf,gBAAgB;gBAChB,cAAc;gBACd,cAAc;aACd,CAAC,CAAC;QACJ,CAAC;QAED,OAAO,MAAM,CAAC;IACf,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,EAAE,CAAC;IACX,CAAC;YAAS,CAAC;QACV,YAAY,CAAC,OAAO,CAAC,CAAC;IACvB,CAAC;AACF,CAAC","sourcesContent":["/**\n * Relay model auto-detection.\n *\n * The relay is just an OpenAI-compatible HTTP endpoint (llama.cpp / vLLM /\n * SGLang style). Hamr treats it as a native provider: at startup it queries\n * the endpoint's `GET /v1/models` and discovers whatever models the relay is\n * currently serving, so the user never has to hand-list them.\n *\n * Ported from synax's `probe-models.ts`: scans the many context-window field\n * names different OpenAI-compatible servers emit, reads vision capability from\n * the `capabilities` array (llama.cpp emits `[\"completion\", \"multimodal\"]`\n * when `--mmproj` is loaded), and derives a readable display name.\n *\n * All failures are swallowed — an unreachable endpoint yields an empty list,\n * never a crash.\n */\n\n/** A model discovered from a relay / OpenAI-compatible endpoint. */\nexport interface DiscoveredRelayModel {\n\tid: string;\n\tdisplayName: string;\n\tcontextWindow?: number;\n\tmaxOutputTokens?: number;\n\tsupportsThinking: boolean;\n\tthinkingLevels: string[];\n\tsupportsVision?: boolean;\n}\n\n// Common field names used by various OpenAI-compatible servers for context window.\n// Order roughly by likelihood: llama.cpp forks (max_context_length), vLLM/SGLang\n// (max_model_len), generic (context_length), and others.\nconst CONTEXT_WINDOW_FIELDS = [\n\t\"max_context_length\",\n\t\"max_model_len\",\n\t\"context_length\",\n\t\"max_total_tokens\",\n\t\"total_tokens_capacity\",\n\t\"n_ctx\",\n\t\"max_position_embeddings\",\n\t\"model_max_length\",\n\t\"max_seq_len\",\n\t\"max_sequence_length\",\n];\n\nconst KNOWN_THINKING_LEVELS = [\"off\", \"on\", \"low\", \"medium\", \"high\", \"xhigh\"];\n\n// Discovery is best-effort and must never block startup for long.\nconst DISCOVERY_TIMEOUT_MS = 5000;\n\nfunction buildHeaders(apiKey?: string, customHeaders?: Record<string, string>): Record<string, string> {\n\tconst headers: Record<string, string> = {\n\t\tAccept: \"application/json\",\n\t\t\"User-Agent\": \"hamr/1.0\",\n\t};\n\tif (apiKey) {\n\t\theaders.Authorization = `Bearer ${apiKey}`;\n\t}\n\tif (customHeaders) {\n\t\tfor (const [key, value] of Object.entries(customHeaders)) {\n\t\t\theaders[key] = value;\n\t\t}\n\t}\n\treturn headers;\n}\n\nfunction extractContextWindow(entry: Record<string, unknown>): number | undefined {\n\t// Check top-level fields first.\n\tfor (const field of CONTEXT_WINDOW_FIELDS) {\n\t\tconst value = entry[field];\n\t\tif (typeof value === \"number\" && Number.isFinite(value) && value > 0) {\n\t\t\treturn value;\n\t\t}\n\t\tif (typeof value === \"string\") {\n\t\t\tconst parsed = Number.parseInt(value, 10);\n\t\t\tif (!Number.isNaN(parsed) && parsed > 0) return parsed;\n\t\t}\n\t}\n\t// Some servers (e.g. llama.cpp) nest context-length fields inside meta.\n\tconst meta = entry.meta as Record<string, unknown> | undefined;\n\tif (meta && typeof meta === \"object\") {\n\t\tfor (const field of CONTEXT_WINDOW_FIELDS) {\n\t\t\tconst value = meta[field];\n\t\t\tif (typeof value === \"number\" && Number.isFinite(value) && value > 0) {\n\t\t\t\treturn value;\n\t\t\t}\n\t\t\tif (typeof value === \"string\") {\n\t\t\t\tconst parsed = Number.parseInt(value, 10);\n\t\t\t\tif (!Number.isNaN(parsed) && parsed > 0) return parsed;\n\t\t\t}\n\t\t}\n\t}\n\treturn undefined;\n}\n\n/**\n * Derive a human-readable display name from a model ID, stripping common\n * file extensions and quantization suffixes (gguf / IQ / Q*_K variants).\n */\nfunction deriveDisplayName(id: string): string {\n\tconst cleaned = id\n\t\t.replace(/\\.gguf$/i, \"\")\n\t\t.replace(/[-_]IQ[23456]_\\w+$/i, \"\")\n\t\t.replace(/[-_]Q[234568]_K_\\w+$/i, \"\")\n\t\t.replace(/[-_](f16|f32|q4_0|q4_1|q5_0|q5_1|q8_0)$/i, \"\");\n\treturn cleaned.replace(/[-_]/g, \" \").replace(/\\b\\w/g, (c) => c.toUpperCase());\n}\n\nfunction stringList(value: unknown): string[] {\n\tif (!Array.isArray(value)) return [];\n\treturn value.map((item) => String(item).toLowerCase());\n}\n\nfunction booleanField(entry: Record<string, unknown>, fields: string[]): boolean | undefined {\n\tfor (const field of fields) {\n\t\tconst value = entry[field];\n\t\tif (typeof value === \"boolean\") return value;\n\t\tif (typeof value === \"string\") {\n\t\t\tconst lower = value.toLowerCase();\n\t\t\tif ([\"true\", \"yes\", \"1\"].includes(lower)) return true;\n\t\t\tif ([\"false\", \"no\", \"0\"].includes(lower)) return false;\n\t\t}\n\t}\n\treturn undefined;\n}\n\nfunction extractSupportsVision(entry: Record<string, unknown>): boolean | undefined {\n\tconst explicit = booleanField(entry, [\n\t\t\"supports_vision\",\n\t\t\"supportsVision\",\n\t\t\"vision\",\n\t\t\"multimodal\",\n\t\t\"supports_images\",\n\t\t\"supportsImages\",\n\t]);\n\tif (explicit !== undefined) return explicit;\n\n\tconst meta = entry.meta && typeof entry.meta === \"object\" ? (entry.meta as Record<string, unknown>) : undefined;\n\tif (meta) {\n\t\tconst metaExplicit = booleanField(meta, [\n\t\t\t\"supports_vision\",\n\t\t\t\"supportsVision\",\n\t\t\t\"vision\",\n\t\t\t\"multimodal\",\n\t\t\t\"supports_images\",\n\t\t\t\"supportsImages\",\n\t\t]);\n\t\tif (metaExplicit !== undefined) return metaExplicit;\n\t}\n\n\tconst haystack = [\n\t\t...stringList(entry.capabilities),\n\t\t...stringList(entry.modalities),\n\t\t...stringList(entry.input_modalities),\n\t\t...stringList(entry.inputModalities),\n\t\t...stringList(entry.features),\n\t\t...(meta ? stringList(meta.capabilities) : []),\n\t\t...(meta ? stringList(meta.modalities) : []),\n\t\t...(meta ? stringList(meta.input_modalities) : []),\n\t\t...(meta ? stringList(meta.inputModalities) : []),\n\t\t...(meta ? stringList(meta.features) : []),\n\t];\n\tif (haystack.some((item) => item === \"multimodal\" || item === \"vision\" || item === \"image\" || item === \"images\")) {\n\t\treturn true;\n\t}\n\tif (haystack.some((item) => item === \"text-only\" || item === \"text_only\")) {\n\t\treturn false;\n\t}\n\treturn undefined;\n}\n\n/**\n * Fetch available models from an OpenAI-compatible `GET /v1/models` endpoint.\n *\n * Returns the discovered models, or an empty array if the endpoint could not\n * be reached or returned an unrecognized payload. Never throws.\n */\nexport async function discoverRelayModels(\n\tbaseUrl: string,\n\tapiKey?: string,\n\tcustomHeaders?: Record<string, string>,\n): Promise<DiscoveredRelayModel[]> {\n\tconst cleanBaseUrl = baseUrl.replace(/\\/+$/, \"\");\n\tconst headers = buildHeaders(apiKey, customHeaders);\n\tconst controller = new AbortController();\n\tconst timeout = setTimeout(() => controller.abort(), DISCOVERY_TIMEOUT_MS);\n\n\ttry {\n\t\tconst res = await fetch(`${cleanBaseUrl}/models`, {\n\t\t\tmethod: \"GET\",\n\t\t\theaders,\n\t\t\tsignal: controller.signal,\n\t\t});\n\t\tif (!res.ok) return [];\n\n\t\tconst body = (await res.json()) as unknown;\n\n\t\t// Standard OpenAI /v1/models response: { object: \"list\", data: [...] }\n\t\t// Some servers return a bare array instead.\n\t\tconst data = (body as { data?: unknown })?.data;\n\t\tconst entries = Array.isArray(data)\n\t\t\t? (data as Array<Record<string, unknown>>)\n\t\t\t: Array.isArray(body)\n\t\t\t\t? (body as Array<Record<string, unknown>>)\n\t\t\t\t: undefined;\n\t\tif (!entries) return [];\n\n\t\tconst models: DiscoveredRelayModel[] = [];\n\t\tfor (const entry of entries) {\n\t\t\tconst rawId = typeof entry.id === \"string\" ? entry.id : typeof entry.name === \"string\" ? entry.name : \"\";\n\t\t\tconst id = rawId.trim();\n\t\t\tif (!id) continue;\n\n\t\t\tconst contextWindow = extractContextWindow(entry);\n\n\t\t\t// Prefer server-provided display_name / name, fall back to deriving from ID.\n\t\t\tconst serverName =\n\t\t\t\t(typeof entry.display_name === \"string\" && entry.display_name.trim()) ||\n\t\t\t\t(typeof entry.name === \"string\" && entry.name.trim()) ||\n\t\t\t\t\"\";\n\t\t\tconst displayName = serverName ? serverName : deriveDisplayName(id);\n\n\t\t\tconst supportsVision = extractSupportsVision(entry);\n\n\t\t\tconst supportsThinking = typeof entry.supports_thinking === \"boolean\" ? entry.supports_thinking : false;\n\t\t\tconst apiLevels = Array.isArray(entry.thinking_levels)\n\t\t\t\t? (entry.thinking_levels as unknown[]).map((l) => String(l)).filter((l) => KNOWN_THINKING_LEVELS.includes(l))\n\t\t\t\t: [];\n\t\t\t// When the server says thinking is supported but doesn't advertise\n\t\t\t// specific levels, default to boolean on/off (standard for local models).\n\t\t\tconst thinkingLevels = apiLevels.length > 0 ? apiLevels : supportsThinking ? [\"off\", \"on\"] : [];\n\n\t\t\tconst maxOutputTokens =\n\t\t\t\ttypeof entry.max_output_tokens === \"number\" && entry.max_output_tokens > 0\n\t\t\t\t\t? entry.max_output_tokens\n\t\t\t\t\t: undefined;\n\n\t\t\tmodels.push({\n\t\t\t\tid,\n\t\t\t\tdisplayName,\n\t\t\t\tcontextWindow,\n\t\t\t\tmaxOutputTokens,\n\t\t\t\tsupportsThinking,\n\t\t\t\tthinkingLevels,\n\t\t\t\tsupportsVision,\n\t\t\t});\n\t\t}\n\n\t\treturn models;\n\t} catch {\n\t\treturn [];\n\t} finally {\n\t\tclearTimeout(timeout);\n\t}\n}\n"]}

package/dist/hamr/providers/repair/json-repair.d.ts

@@ -0,0 +1,30 @@
+/**
+ * JSON repair — bounded auto-recovery for local-model tool-call JSON.
+ *
+ * Local models frequently emit malformed JSON with:
+ * - Trailing commas: {"name": "read", "args": {"path": "x",}}
+ * - Unescaped inner quotes: {"query": "find "foo" in bar"}
+ * - Truncated objects: {"name": "bash", "args": {"command": "npm te
+ * - Missing closing braces: {"name": "edit", "args": {"path": "x", "oldStr": "y"
+ * - Mixed format: prose + partial JSON blocks
+ *
+ * Each repair is recorded in `fixes[]` for debugging. Returns `null` when
+ * the input is unrepairable (garbage, empty, or too broken to guess).
+ */
+export interface RepairResult {
+    repaired: string;
+    fixes: string[];
+}
+/**
+ * Attempt to repair malformed JSON produced by a local model.
+ *
+ * Repairs are applied in increasing order of invasiveness:
+ * 1. Trim whitespace and surrounding noise
+ * 2. Fix trailing commas
+ * 3. Balance braces/brackets
+ * 4. Heuristic inner-quote repair
+ *
+ * Returns `null` if the string is unrepairable.
+ */
+export declare function repairJson(raw: string): RepairResult | null;
+//# sourceMappingURL=json-repair.d.ts.map

package/dist/hamr/providers/repair/json-repair.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"json-repair.d.ts","sourceRoot":"","sources":["../../../../src/hamr/providers/repair/json-repair.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,MAAM,WAAW,YAAY;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,EAAE,CAAC;CAChB;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,UAAU,CAAC,GAAG,EAAE,MAAM,GAAG,YAAY,GAAG,IAAI,CAqD3D"}