@arbidocs/client 0.3.177 → 0.3.179

package/dist/index.d.cts

@@ -4124,7 +4124,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are ARBI, an AI assistant created by ARBI CITY. Be proactive, helpful and professional.Be concise. Use markdown where it helps readability. Cross-check information across sources before answering. Never fabricate facts. Not all assertions in the sources are factual — some are arguments by a party. Attribute claims to their source, especially where parties disagree.",
              *       "MAX_TOKEN_SIZE_TO_ANSWER": 200000,
              *       "TEMPERATURE": 0.1,
@@ -4136,7 +4136,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are reviewing a draft answer prepared by a research agent. The draft was written based on summaries, but you now have access to the full source material.\n\nYour task:\n1. Review the draft answer against the source material provided\n2. Verify claims are supported by the sources\n3. Correct any inaccuracies or unsupported statements\n4. Add any important details from the sources that were missed\n5. Maintain formal, objective tone appropriate for professional contexts\n\nIf information is not found in the sources but the draft answer is based on web search results or general knowledge, preserve it and ensure it is clearly labelled as such. Only flag missing source support for claims that purport to come from the documents.\nDo not add inline citation markers - the system handles citations automatically.",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
@@ -4148,7 +4148,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a chunk evaluator. Analyze retrieved chunks to determine which are relevant to the user's query and extract key learnings from the RELEVANT chunks only. You must return your response as valid JSON matching the provided schema.",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
@@ -4163,7 +4163,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a title generator that creates concise, descriptive titles.\n\nOutput the title as plain text only. Do not use Markdown or any other formatting. Specifically:\n- Do not bold, italicize, or otherwise emphasize any part of the title.\n- Do not include asterisks (*), underscores (_), backticks (`), or other Markdown markers.\n- Do not prefix the title with bullets, dashes (-), or numbered/lettered enumeration (e.g. \"1.\", \"a)\").\n- Do not wrap the title in quotes or brackets.",
              *       "MAX_TOKEN_SIZE_TO_ANSWER": 20,
              *       "TEMPERATURE": 0.3,
@@ -4175,7 +4175,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a conversation summariser. Condense the conversation history into a concise summary that preserves:\n- Key decisions and conclusions reached\n- Specific names, dates, numbers, and document references\n- Open questions and unresolved items\n- The user's goals and constraints\n\nWrite in past tense, third person. Be specific, not vague. Do not omit important details in favour of brevity.",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
@@ -4189,7 +4189,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a document analysis assistant. Answer questions based strictly on document content.\n\nYour answers are displayed directly as cell values in a table, so keep every answer as short as possible.\n\nGuidelines:\n- Answer with the value only — never a full sentence. Do not restate the question or add lead-ins such as \"This document is...\" or \"The document describes...\". Good: \"German\". Bad: \"This document is written in German.\"\n- When an answer has multiple values, put each value on its own line: separate values with a newline character. Do not separate them with commas, bullets, dashes, numbering, or by writing them as a sentence.\n- Only use information explicitly stated in the document; do not infer or assume information not present.\n- If the information is not found, respond with exactly \"Not found\".\n- Follow the format instructions exactly for each answer.\n- Output answers as plain text only. Do not use Markdown or any other formatting: no bold, italics, asterisks (*), underscores (_), backticks (`), bullets, or numbered/lettered enumeration.",
              *       "MAX_TOKEN_CONTEXT_TO_ANSWER": 200000,
              *       "TEMPERATURE": 0.1,
@@ -4250,7 +4250,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "TEMPERATURE": 0.3,
              *       "MAX_TOKENS": 0,
              *       "MAX_TOKEN_CONTEXT": 200000,
@@ -4262,7 +4262,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": true,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a research planning assistant. Your job is to analyze a user query and produce a concise, numbered research plan.\n\nYou have access to a workspace of documents. The available tools for executing the plan are:\n- search_inside_documents: Search workspace documents for relevant passages (semantic, keyword, or hybrid). Use digest=true to extract learnings.\n- read_document: Read page ranges or entire documents. Use digest=true to extract learnings with citations.\n- get_table_of_contents: Get document headings with page references\n\nConsider:\n- What information is needed to answer the query\n- Which documents are likely relevant based on the document index\n- What search queries and document passages to examine\n- What order of operations will be most efficient\n- Whether parallel searches can be used for different aspects\n\nRespond with ONLY the plan (numbered steps). Do not execute any steps.",
              *       "TEMPERATURE": 0.3,
              *       "MAX_TOKENS": 0,
@@ -4275,7 +4275,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a context selector. Given a user's query, conversation history, and a table of available documents, determine the user's goal and select relevant context.\n\n## Instructions\n\n1. **goal**: 1-2 concise sentences describing the user's intent AND your filtering rationale. Explain what documents are needed and why. Examples:\n   - \"The user is greeting the assistant. No documents are needed.\"\n   - \"The user wants to find force majeure clauses. Only the three contracts are relevant, so I will keep only those.\"\n   - \"The user wants a full timeline of the case. Most documents are relevant, so I will drop only the unrelated procedural orders.\"\n\n2. **doc_filter_mode** and **doc_ids**: Select which documents the agent needs.\n   - `\"keep\"` mode: `doc_ids` lists the ONLY docs to keep. Everything else is dropped.\n   - `\"drop\"` mode: `doc_ids` lists the ONLY docs to drop. Everything else is kept.\n   - Pick whichever mode produces the **shorter `doc_ids` list**.\n   - Doc IDs are in the **Id** column of the Available Documents table (e.g. `\"doc-abc12345\"`). Copy them **exactly** from the table — do not invent, guess, or modify IDs.\n\n   | Scenario | doc_filter_mode | doc_ids |\n   |----------|----------------|---------|\n   | Need ALL docs | `\"drop\"` | `[]` |\n   | Need NO docs | `\"keep\"` | `[]` |\n   | Need 3 of 100 docs | `\"keep\"` | `[\"id1\",\"id2\",\"id3\"]` |\n   | Drop 3 of 100 docs | `\"drop\"` | `[\"id1\",\"id2\",\"id3\"]` |\n\n3. **chat_range_start** and **chat_range_end**: 0-based inclusive range of earlier history messages that provide relevant context. The last 3 messages are always included automatically. Set both to -1 if no earlier history is needed.\n\n## Rules\n\n- **Never repeat a doc ID.** Each ID must appear at most once.\n- Before writing `doc_ids`, count how many you would keep vs drop, then pick the shorter list.\n- Greetings or meta-questions about the assistant (e.g. \"hello\", \"who are you?\", \"what can you do?\"): the agent does NOT need any documents → `\"keep\"` with `[]`.\n- Ambiguous references like \"the document\" or \"this contract\" without specifying which: keep all → `\"drop\"` with `[]`.\n- For follow-up questions, include history messages that establish the topic.\n- Always return valid JSON matching the schema.",
              *       "TEMPERATURE": 0,
              *       "MAX_TOKENS": 0
@@ -4286,7 +4286,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
              *       "MAX_PAGES_PER_CALL": 5,
@@ -4307,7 +4307,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": true,
-             *       "MODEL_NAME": "Premium",
+             *       "MODEL_NAME": "",
              *       "SYSTEM_INSTRUCTION": "You are a code execution agent. You write and run code to accomplish tasks.\n\nYou have access to an execute_code tool that runs code in a sandboxed Docker container.\n\nThe sandbox has:\n- Python 3.12 with numpy, pandas, matplotlib, requests, beautifulsoup4,\n  httpx, pillow, sympy, pyyaml (plus uv for installing more)\n- Node.js 22 with TypeScript (ts-node) and npm\n- Bash with git, curl, wget, jq, sqlite3, build-essential\n- Network access (can pip/npm install additional packages)\n\nInstructions:\n1. Write clean, correct code to accomplish the user's task\n2. Always call execute_code to run your code — never guess the output\n3. If execution errors, analyze the traceback, fix the code, and retry\n4. Once you have the correct output, respond with ONLY the final result\n\nKeep code simple and direct. Prefer Python unless the task specifically requires another language.",
              *       "TEMPERATURE": 0.2,
              *       "MAX_TOKENS": 0
@@ -4377,7 +4377,7 @@ interface components {
              *       },
              *       "MAX_CONCURRENT_REQUESTS": 256,
              *       "MODEL_NAME": "Qwen/Qwen3-Reranker-0.6B",
-             *       "API_TYPE": "remote",
+             *       "API_TYPE": "local",
              *       "RETRIEVAL_INSTRUCTION": "Judge whether the passage contains the specific facts, figures, names, or references mentioned in the query. Only answer yes if the passage directly contains these details"
              *     }
              */
@@ -4398,7 +4398,7 @@ interface components {
             /**
              * @default {
              *       "MODEL_NAME": "Qwen/Qwen3-Embedding-0.6B",
-             *       "API_TYPE": "remote",
+             *       "API_TYPE": "local",
              *       "BATCH_SIZE": 128,
              *       "MAX_CONCURRENT_REQUESTS": 256,
              *       "DOCUMENT_PREFIX": "",
@@ -5050,49 +5050,49 @@ interface components {
         };
         /**
          * CitationSources
-         * @description Flat provenance for a grounded note: the passages that back it.
-         *
-         *     A note or a DocTag answers ONE question, so the per-statement
-         *     :class:`CitationData` map it used to carry was almost entirely redundant —
-         *     the ``statement`` duplicated the note text and the offsets spanned it end to
-         *     end. All a reader needs is *which passages support this*, best-first.
+         * @description Provenance for a grounded note: the documents it draws on, best first,
+         *     each with the passages cited from it, best first within the document.
          *
-         *     One object per passage, deliberately, rather than parallel arrays: document
-         *     and page belong to a specific chunk, and keeping them together removes any
-         *     alignment invariant to maintain. Everything a client renders derives from
-         *     this single list — document pills by deduping ``doc_ext_id``, page ranges by
-         *     grouping on it — so there is exactly one way to express provenance.
+         *     Grouped by document because that is the unit every reader works in — the
+         *     Sources pills, the picker checkbox, "remove this source" — and because it
+         *     lets a whole-document source be expressed as a document with no chunks
+         *     rather than a sentinel chunk id. Stored as a bare JSON array.
          *
-         *     Rerank scores are NOT stored. Nothing ever read them back, and a score is
-         *     not comparable across reranker versions, so persisting one would imply a
-         *     precision it does not have. Order carries the only durable signal.
+         *     Rerank scores are NOT stored: nothing ever read them back and they are not
+         *     comparable across reranker versions. Order carries the only durable signal.
          *
-         *     Holds no document text, so it is **plaintext at rest** — chunk and doc ext
-         *     ids are opaque identifiers, and the passage content behind them stays
-         *     SecretBox-encrypted in Qdrant.
+         *     Holds no document text — chunk and doc ext ids are opaque — so it is
+         *     **plaintext at rest**; the passage content stays SecretBox-encrypted in
+         *     Qdrant. ``None`` (never ``[]``) is the one representation of "ungrounded".
          */
-        CitationSources: {
-            /** Passages */
-            passages: components['schemas']['CitedPassage'][];
-        };
+        CitationSources: components['schemas']['CitedDocument'][];
         /**
-         * CitedPassage
-         * @description One passage backing a grounded note, with enough to render it.
-         *
-         *     ``doc_ext_id`` is captured at grounding time from the chunk's metadata so a
-         *     client can show the source as a document pill without resolving every chunk
-         *     against Qdrant first. It is optional only because citations written before
-         *     that capture existed carry the chunk alone — a missing value means "not
-         *     recorded", never "no document" (every chunk belongs to exactly one).
+         * CitedChunk
+         * @description One passage backing a grounded note: the chunk, and the page it sits on
+         *     (captured at grounding time so a client can show a page pill without
+         *     resolving the chunk against Qdrant).
          */
-        CitedPassage: {
+        CitedChunk: {
             /** Chunk Id */
             chunk_id: string;
-            /** Doc Ext Id */
-            doc_ext_id?: string | null;
             /** Page */
             page?: number | null;
         };
+        /**
+         * CitedDocument
+         * @description One document backing a grounded note, with the passages cited from it.
+         *
+         *     An EMPTY ``chunks`` list means the whole document is the source — a user
+         *     ticked it in the picker without choosing passages. There is no separate
+         *     "document-only" field: the document is always the unit, and passages
+         *     narrow it.
+         */
+        CitedDocument: {
+            /** Doc Ext Id */
+            doc_ext_id: string;
+            /** Chunks */
+            chunks?: components['schemas']['CitedChunk'][];
+        };
         /**
          * CodeAgentConfig
          * @description Configuration for CodeAgent — sub-agent that writes and executes code via sandbox.
@@ -5114,7 +5114,7 @@ interface components {
             /**
              * Model Name
              * @description Model for code generation. Should be a strong coding model.
-             * @default Premium
+             * @default
              */
             MODEL_NAME: string;
             /**
@@ -6298,7 +6298,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -6557,7 +6557,7 @@ interface components {
             /**
              * Api Type
              * @description The inference type (local or remote).
-             * @default remote
+             * @default local
              * @enum {string}
              */
             API_TYPE: 'local' | 'remote';
@@ -6697,7 +6697,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -6952,7 +6952,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -7561,7 +7561,7 @@ interface components {
             /**
              * Model Name
              * @description The model for note consolidation. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -8350,7 +8350,7 @@ interface components {
             /**
              * Model Name
              * @description The model for plan generation. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -8743,7 +8743,7 @@ interface components {
             /**
              * Model Name
              * @description The model for query answering.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -9068,7 +9068,7 @@ interface components {
             /**
              * Api Type
              * @description The inference type (local or remote).
-             * @default remote
+             * @default local
              * @enum {string}
              */
             API_TYPE: 'local' | 'remote';
@@ -9851,7 +9851,7 @@ interface components {
             /**
              * Model Name
              * @description The model for reviewing agent draft answers. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -10425,7 +10425,7 @@ interface components {
             /**
              * Model Name
              * @description The model for conversation summarisation. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -10588,6 +10588,11 @@ interface components {
             parent_ext_id?: string | null;
             /** Doctag Count */
             doctag_count: number;
+            /**
+             * Facttag Count
+             * @default 0
+             */
+            facttag_count: number;
             /** Created By Ext Id */
             created_by_ext_id: string;
             /** Updated By Ext Id */
@@ -10847,7 +10852,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -11784,7 +11789,7 @@ interface components {
             /**
              * Model Name
              * @description The vision-capable model name (must support image inputs).
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**

package/dist/index.d.ts

@@ -4124,7 +4124,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are ARBI, an AI assistant created by ARBI CITY. Be proactive, helpful and professional.Be concise. Use markdown where it helps readability. Cross-check information across sources before answering. Never fabricate facts. Not all assertions in the sources are factual — some are arguments by a party. Attribute claims to their source, especially where parties disagree.",
              *       "MAX_TOKEN_SIZE_TO_ANSWER": 200000,
              *       "TEMPERATURE": 0.1,
@@ -4136,7 +4136,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are reviewing a draft answer prepared by a research agent. The draft was written based on summaries, but you now have access to the full source material.\n\nYour task:\n1. Review the draft answer against the source material provided\n2. Verify claims are supported by the sources\n3. Correct any inaccuracies or unsupported statements\n4. Add any important details from the sources that were missed\n5. Maintain formal, objective tone appropriate for professional contexts\n\nIf information is not found in the sources but the draft answer is based on web search results or general knowledge, preserve it and ensure it is clearly labelled as such. Only flag missing source support for claims that purport to come from the documents.\nDo not add inline citation markers - the system handles citations automatically.",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
@@ -4148,7 +4148,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a chunk evaluator. Analyze retrieved chunks to determine which are relevant to the user's query and extract key learnings from the RELEVANT chunks only. You must return your response as valid JSON matching the provided schema.",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
@@ -4163,7 +4163,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a title generator that creates concise, descriptive titles.\n\nOutput the title as plain text only. Do not use Markdown or any other formatting. Specifically:\n- Do not bold, italicize, or otherwise emphasize any part of the title.\n- Do not include asterisks (*), underscores (_), backticks (`), or other Markdown markers.\n- Do not prefix the title with bullets, dashes (-), or numbered/lettered enumeration (e.g. \"1.\", \"a)\").\n- Do not wrap the title in quotes or brackets.",
              *       "MAX_TOKEN_SIZE_TO_ANSWER": 20,
              *       "TEMPERATURE": 0.3,
@@ -4175,7 +4175,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a conversation summariser. Condense the conversation history into a concise summary that preserves:\n- Key decisions and conclusions reached\n- Specific names, dates, numbers, and document references\n- Open questions and unresolved items\n- The user's goals and constraints\n\nWrite in past tense, third person. Be specific, not vague. Do not omit important details in favour of brevity.",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
@@ -4189,7 +4189,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a document analysis assistant. Answer questions based strictly on document content.\n\nYour answers are displayed directly as cell values in a table, so keep every answer as short as possible.\n\nGuidelines:\n- Answer with the value only — never a full sentence. Do not restate the question or add lead-ins such as \"This document is...\" or \"The document describes...\". Good: \"German\". Bad: \"This document is written in German.\"\n- When an answer has multiple values, put each value on its own line: separate values with a newline character. Do not separate them with commas, bullets, dashes, numbering, or by writing them as a sentence.\n- Only use information explicitly stated in the document; do not infer or assume information not present.\n- If the information is not found, respond with exactly \"Not found\".\n- Follow the format instructions exactly for each answer.\n- Output answers as plain text only. Do not use Markdown or any other formatting: no bold, italics, asterisks (*), underscores (_), backticks (`), bullets, or numbered/lettered enumeration.",
              *       "MAX_TOKEN_CONTEXT_TO_ANSWER": 200000,
              *       "TEMPERATURE": 0.1,
@@ -4250,7 +4250,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "TEMPERATURE": 0.3,
              *       "MAX_TOKENS": 0,
              *       "MAX_TOKEN_CONTEXT": 200000,
@@ -4262,7 +4262,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": true,
-             *       "MODEL_NAME": "Wise",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a research planning assistant. Your job is to analyze a user query and produce a concise, numbered research plan.\n\nYou have access to a workspace of documents. The available tools for executing the plan are:\n- search_inside_documents: Search workspace documents for relevant passages (semantic, keyword, or hybrid). Use digest=true to extract learnings.\n- read_document: Read page ranges or entire documents. Use digest=true to extract learnings with citations.\n- get_table_of_contents: Get document headings with page references\n\nConsider:\n- What information is needed to answer the query\n- Which documents are likely relevant based on the document index\n- What search queries and document passages to examine\n- What order of operations will be most efficient\n- Whether parallel searches can be used for different aspects\n\nRespond with ONLY the plan (numbered steps). Do not execute any steps.",
              *       "TEMPERATURE": 0.3,
              *       "MAX_TOKENS": 0,
@@ -4275,7 +4275,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "SYSTEM_INSTRUCTION": "You are a context selector. Given a user's query, conversation history, and a table of available documents, determine the user's goal and select relevant context.\n\n## Instructions\n\n1. **goal**: 1-2 concise sentences describing the user's intent AND your filtering rationale. Explain what documents are needed and why. Examples:\n   - \"The user is greeting the assistant. No documents are needed.\"\n   - \"The user wants to find force majeure clauses. Only the three contracts are relevant, so I will keep only those.\"\n   - \"The user wants a full timeline of the case. Most documents are relevant, so I will drop only the unrelated procedural orders.\"\n\n2. **doc_filter_mode** and **doc_ids**: Select which documents the agent needs.\n   - `\"keep\"` mode: `doc_ids` lists the ONLY docs to keep. Everything else is dropped.\n   - `\"drop\"` mode: `doc_ids` lists the ONLY docs to drop. Everything else is kept.\n   - Pick whichever mode produces the **shorter `doc_ids` list**.\n   - Doc IDs are in the **Id** column of the Available Documents table (e.g. `\"doc-abc12345\"`). Copy them **exactly** from the table — do not invent, guess, or modify IDs.\n\n   | Scenario | doc_filter_mode | doc_ids |\n   |----------|----------------|---------|\n   | Need ALL docs | `\"drop\"` | `[]` |\n   | Need NO docs | `\"keep\"` | `[]` |\n   | Need 3 of 100 docs | `\"keep\"` | `[\"id1\",\"id2\",\"id3\"]` |\n   | Drop 3 of 100 docs | `\"drop\"` | `[\"id1\",\"id2\",\"id3\"]` |\n\n3. **chat_range_start** and **chat_range_end**: 0-based inclusive range of earlier history messages that provide relevant context. The last 3 messages are always included automatically. Set both to -1 if no earlier history is needed.\n\n## Rules\n\n- **Never repeat a doc ID.** Each ID must appear at most once.\n- Before writing `doc_ids`, count how many you would keep vs drop, then pick the shorter list.\n- Greetings or meta-questions about the assistant (e.g. \"hello\", \"who are you?\", \"what can you do?\"): the agent does NOT need any documents → `\"keep\"` with `[]`.\n- Ambiguous references like \"the document\" or \"this contract\" without specifying which: keep all → `\"drop\"` with `[]`.\n- For follow-up questions, include history messages that establish the topic.\n- Always return valid JSON matching the schema.",
              *       "TEMPERATURE": 0,
              *       "MAX_TOKENS": 0
@@ -4286,7 +4286,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": false,
-             *       "MODEL_NAME": "Fast",
+             *       "MODEL_NAME": "dummy",
              *       "TEMPERATURE": 0.1,
              *       "MAX_TOKENS": 0,
              *       "MAX_PAGES_PER_CALL": 5,
@@ -4307,7 +4307,7 @@ interface components {
              * @default {
              *       "API_TYPE": "remote",
              *       "ENABLE_THINKING": true,
-             *       "MODEL_NAME": "Premium",
+             *       "MODEL_NAME": "",
              *       "SYSTEM_INSTRUCTION": "You are a code execution agent. You write and run code to accomplish tasks.\n\nYou have access to an execute_code tool that runs code in a sandboxed Docker container.\n\nThe sandbox has:\n- Python 3.12 with numpy, pandas, matplotlib, requests, beautifulsoup4,\n  httpx, pillow, sympy, pyyaml (plus uv for installing more)\n- Node.js 22 with TypeScript (ts-node) and npm\n- Bash with git, curl, wget, jq, sqlite3, build-essential\n- Network access (can pip/npm install additional packages)\n\nInstructions:\n1. Write clean, correct code to accomplish the user's task\n2. Always call execute_code to run your code — never guess the output\n3. If execution errors, analyze the traceback, fix the code, and retry\n4. Once you have the correct output, respond with ONLY the final result\n\nKeep code simple and direct. Prefer Python unless the task specifically requires another language.",
              *       "TEMPERATURE": 0.2,
              *       "MAX_TOKENS": 0
@@ -4377,7 +4377,7 @@ interface components {
              *       },
              *       "MAX_CONCURRENT_REQUESTS": 256,
              *       "MODEL_NAME": "Qwen/Qwen3-Reranker-0.6B",
-             *       "API_TYPE": "remote",
+             *       "API_TYPE": "local",
              *       "RETRIEVAL_INSTRUCTION": "Judge whether the passage contains the specific facts, figures, names, or references mentioned in the query. Only answer yes if the passage directly contains these details"
              *     }
              */
@@ -4398,7 +4398,7 @@ interface components {
             /**
              * @default {
              *       "MODEL_NAME": "Qwen/Qwen3-Embedding-0.6B",
-             *       "API_TYPE": "remote",
+             *       "API_TYPE": "local",
              *       "BATCH_SIZE": 128,
              *       "MAX_CONCURRENT_REQUESTS": 256,
              *       "DOCUMENT_PREFIX": "",
@@ -5050,49 +5050,49 @@ interface components {
         };
         /**
          * CitationSources
-         * @description Flat provenance for a grounded note: the passages that back it.
-         *
-         *     A note or a DocTag answers ONE question, so the per-statement
-         *     :class:`CitationData` map it used to carry was almost entirely redundant —
-         *     the ``statement`` duplicated the note text and the offsets spanned it end to
-         *     end. All a reader needs is *which passages support this*, best-first.
+         * @description Provenance for a grounded note: the documents it draws on, best first,
+         *     each with the passages cited from it, best first within the document.
          *
-         *     One object per passage, deliberately, rather than parallel arrays: document
-         *     and page belong to a specific chunk, and keeping them together removes any
-         *     alignment invariant to maintain. Everything a client renders derives from
-         *     this single list — document pills by deduping ``doc_ext_id``, page ranges by
-         *     grouping on it — so there is exactly one way to express provenance.
+         *     Grouped by document because that is the unit every reader works in — the
+         *     Sources pills, the picker checkbox, "remove this source" — and because it
+         *     lets a whole-document source be expressed as a document with no chunks
+         *     rather than a sentinel chunk id. Stored as a bare JSON array.
          *
-         *     Rerank scores are NOT stored. Nothing ever read them back, and a score is
-         *     not comparable across reranker versions, so persisting one would imply a
-         *     precision it does not have. Order carries the only durable signal.
+         *     Rerank scores are NOT stored: nothing ever read them back and they are not
+         *     comparable across reranker versions. Order carries the only durable signal.
          *
-         *     Holds no document text, so it is **plaintext at rest** — chunk and doc ext
-         *     ids are opaque identifiers, and the passage content behind them stays
-         *     SecretBox-encrypted in Qdrant.
+         *     Holds no document text — chunk and doc ext ids are opaque — so it is
+         *     **plaintext at rest**; the passage content stays SecretBox-encrypted in
+         *     Qdrant. ``None`` (never ``[]``) is the one representation of "ungrounded".
          */
-        CitationSources: {
-            /** Passages */
-            passages: components['schemas']['CitedPassage'][];
-        };
+        CitationSources: components['schemas']['CitedDocument'][];
         /**
-         * CitedPassage
-         * @description One passage backing a grounded note, with enough to render it.
-         *
-         *     ``doc_ext_id`` is captured at grounding time from the chunk's metadata so a
-         *     client can show the source as a document pill without resolving every chunk
-         *     against Qdrant first. It is optional only because citations written before
-         *     that capture existed carry the chunk alone — a missing value means "not
-         *     recorded", never "no document" (every chunk belongs to exactly one).
+         * CitedChunk
+         * @description One passage backing a grounded note: the chunk, and the page it sits on
+         *     (captured at grounding time so a client can show a page pill without
+         *     resolving the chunk against Qdrant).
          */
-        CitedPassage: {
+        CitedChunk: {
             /** Chunk Id */
             chunk_id: string;
-            /** Doc Ext Id */
-            doc_ext_id?: string | null;
             /** Page */
             page?: number | null;
         };
+        /**
+         * CitedDocument
+         * @description One document backing a grounded note, with the passages cited from it.
+         *
+         *     An EMPTY ``chunks`` list means the whole document is the source — a user
+         *     ticked it in the picker without choosing passages. There is no separate
+         *     "document-only" field: the document is always the unit, and passages
+         *     narrow it.
+         */
+        CitedDocument: {
+            /** Doc Ext Id */
+            doc_ext_id: string;
+            /** Chunks */
+            chunks?: components['schemas']['CitedChunk'][];
+        };
         /**
          * CodeAgentConfig
          * @description Configuration for CodeAgent — sub-agent that writes and executes code via sandbox.
@@ -5114,7 +5114,7 @@ interface components {
             /**
              * Model Name
              * @description Model for code generation. Should be a strong coding model.
-             * @default Premium
+             * @default
              */
             MODEL_NAME: string;
             /**
@@ -6298,7 +6298,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -6557,7 +6557,7 @@ interface components {
             /**
              * Api Type
              * @description The inference type (local or remote).
-             * @default remote
+             * @default local
              * @enum {string}
              */
             API_TYPE: 'local' | 'remote';
@@ -6697,7 +6697,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -6952,7 +6952,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -7561,7 +7561,7 @@ interface components {
             /**
              * Model Name
              * @description The model for note consolidation. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -8350,7 +8350,7 @@ interface components {
             /**
              * Model Name
              * @description The model for plan generation. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -8743,7 +8743,7 @@ interface components {
             /**
              * Model Name
              * @description The model for query answering.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -9068,7 +9068,7 @@ interface components {
             /**
              * Api Type
              * @description The inference type (local or remote).
-             * @default remote
+             * @default local
              * @enum {string}
              */
             API_TYPE: 'local' | 'remote';
@@ -9851,7 +9851,7 @@ interface components {
             /**
              * Model Name
              * @description The model for reviewing agent draft answers. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -10425,7 +10425,7 @@ interface components {
             /**
              * Model Name
              * @description The model for conversation summarisation. Defaults to reasoning model.
-             * @default Wise
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -10588,6 +10588,11 @@ interface components {
             parent_ext_id?: string | null;
             /** Doctag Count */
             doctag_count: number;
+            /**
+             * Facttag Count
+             * @default 0
+             */
+            facttag_count: number;
             /** Created By Ext Id */
             created_by_ext_id: string;
             /** Updated By Ext Id */
@@ -10847,7 +10852,7 @@ interface components {
             /**
              * Model Name
              * @description The name of the non-reasoning model to be used.
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**
@@ -11784,7 +11789,7 @@ interface components {
             /**
              * Model Name
              * @description The vision-capable model name (must support image inputs).
-             * @default Fast
+             * @default dummy
              */
             MODEL_NAME: string;
             /**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@arbidocs/client",
-  "version": "0.3.177",
+  "version": "0.3.179",
   "description": "TypeScript SDK for the ARBI API — zero-knowledge auth, E2E encryption, and type-safe REST client",
   "type": "module",
   "main": "dist/index.cjs",