Advanced environmental variables
This page lists the internal behavior of the variable control system (parallel, repeat, overtime, batch size, etc.). ** All the default values are used when not set, and the behavior is completely consistent with the official release. ** Adjust only when the impact is clearly understood; the core deployment variables are seen. Core environmental variables。
0 Overview
共 301 Environmental variables, the sub-system distribution:
| The subsystem | Number of variables |
|---|---|
| LLM Conversation / Arrangement / Internal Model Call | 65 |
| RAG Document Analysis/Vector Recovery | 59 |
| Execution of sandbox code | 9 |
| Built-in tools (Search / Python / Cyber Security) | 14 |
| Session / Message / Streaming API | 73 |
| Certification / Session / Verification Code | 15 |
| Upload / Documentation / Sharing | 14 |
| Admin console tasks (backup / vector maintenance / redeem code) | 20 |
| Server Launch / Configuration | 3 |
| 10 The front | 23 |
| ** Total of ** | 295 |
LLM Conversation / Arrangement / Internal Model Call
Main conversation streams, tool cycles, TFT watch dogs, and internal model calls such as compression/memory/audit/examination/deep research/document generation.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_LLM_APPLY_ANTHROPIC_THINKING_SETTINGS | int | 2048 | When raising the Anthropic max_tokens, extend the token balance added above the budget_tokens to enable the output to accommodate thought plus response. |
AIVORY_LLM_MAX_ITER | int | 20 | The hard threshold for native tool call rounds (Messages API calls) in the Anthropic Stream cycle. |
AIVORY_LLM_MAX_TOK | int | 64000 | The default max_tokens sent for each Anthropic Messages request in the streaming tool cycle (unless the request itself covers). |
AIVORY_LLM_MAX_TOK_2 | int | 64000 | PromptRunOnce is the default max_tokens for a single Anthropic call (unless the request is covered). |
AIVORY_LLM_INFLIGHT_GRACE | duration | 15*time.Minute | The length of time the assistant message line for status=streaming is exempt from being processed in the summary before it is considered a collapse residual. |
AIVORY_LLM_T | int | 4 | The structural token expenditure added for each message in the compressed token estimate (role tags, outboxes). |
AIVORY_LLM_MESSAGE_TOKEN_MEMO_CACHE_BOUND | int | 100000 | Each message token estimates the maximum number of entries allowed before memory cache maps are reset to limit memory. |
AIVORY_LLM_SUMMARY_TOKENS_CLAMP_FLOOR | int | 256 | Manages the bottom limit of the summary_max_tokens setting; resets the default summary budget when the configuration value is less than that bottom limit. |
AIVORY_LLM_BIG_TOKEN_OVERFLOW_NUM | int | 5 | This round triggers the inbound summary token triggering the threshold multiplier (num/den, 5/4 = 1.25 times by default). |
AIVORY_LLM_BIG_TOKEN_OVERFLOW_DEN | int | 4 | This round triggers the inbound summary token trigger the threshold multiplier (num/den, 5/4 = 1.25 times by default). |
AIVORY_LLM_INLINE_COMPACTION_BACKLOG_FACTOR | int | 3 | There is no summary of the multiplier of the tail degree relative to keepRounds*2, exceeding which is compulsory in connection (rather than asynchronous) compression. |
AIVORY_LLM_SUMMARY_MERGE_BUDGET | int | 2048 | It triggers the folding of old summary blocks into more rough cumulative summary token thresholds. |
AIVORY_LLM_DETERMINISTIC_SUMMARY_CLIP_BUDGET | int | 300 | When the task model summary returns as empty, the token budget used to cut backward for the older rounds. |
AIVORY_LLM_ATTEMPT | int | 4 | The maximum number of comparisons and exchanges (CAS) repeated attempts when adding a new summary block to the session summary_blocks. |
AIVORY_LLM_ITER | int | 3 | To compress the summary token of a session path back to the maximum number of repeat folding iterations executed within the budget. |
AIVORY_LLM_MAX_OUTPUT_TOKENS_5 | int | 2 | It acts as a combination budget to derive fold-called MaxOutputTokens and a certain cut length deduction. |
AIVORY_LLM_CHUNK_SIZE | int | 400 | Re-validate the size of the message ID batch for each SQL IN(...) query when the summary message still exists (used to circumvent the driver positioner upper limit partition). |
AIVORY_LLM_DR_MAX_ROUNDS | int | 4 | The Deep Research engine runs a search re-validation hard limit for the number of rounds. |
AIVORY_LLM_DR_QUERIES_PER_ROUND | int | 6 | The maximum number of search queries distributed per Deep Research round. |
AIVORY_LLM_DR_FETCH_PER_ROUND | int | 5 | Each Deep Research round selects and reads the maximum number of new source candidates. |
AIVORY_LLM_DR_MIN_DEEP_READS | int | 5 | Deep Research achieves the minimum number of deep-reading sources to be reached before the end of permission (even if the coverage gap seems sufficient). |
AIVORY_LLM_DR_SEARCH_TOP_K | int | 8 | The number of results (top_k) for each Deep Research search. |
AIVORY_LLM_DR_WALL_CLOCK | duration | 5*time.Minute | Limit the total wall clock overtime running throughout the Deep Research engine. |
AIVORY_LLM_DR_CALL_TIMEOUT | duration | 30*time.Second | Deep Research searches or scraps every single call overtime. |
AIVORY_LLM_DEEP_RESEARCH_VALIDATE_TIMEOUT | duration | 75*time.Second | Limit Deep Research’s overtime to the validate phase of weak/single-source arguments reviewed before writing. |
AIVORY_LLM_SCORE_A | float | 9 | Rankings added to candidate sources with URL credibility rating A (reliability dominates in the ranking). |
AIVORY_LLM_SCORE_B | float | 6 | Added ranking points for candidate sources with URL credibility rating B. |
AIVORY_LLM_SCORE_C | float | 3 | Added ranking points for candidate sources with URL credibility grade C. |
AIVORY_LLM_SCORE_KW | float | 1 | The ranking points added by one problem key word (longer than 3) in the candidate source title or summary. |
AIVORY_LLM_SCORE_FRESH_DOMAIN | float | 2 | Added ranking points for candidate sources of domain names that have not yet been seen in this study. |
AIVORY_LLM_MAX_ITER_4 | int | 20 | In the Gemini Stream cycle, there is a hard upper limit for the number of native tool calls (generateContent calls). |
AIVORY_LLM_GEMINI_MAX_TOK | int | 64000 | The default value for generationConfig.maxOutputTokens is requested each time in the Gemini stream cycle (unless the request is covered). |
AIVORY_LLM_GEMINI_MAX_TOK_2 | int | 64000 | The default value for generationConfig.maxOutputTokens is called in the Gemini Tipword Tool mode (unless the request is covered). |
AIVORY_LLM_CONF | float | 0.7 | When the extractor returns a credibility not within the range (0,1), the extractor returns the credibility assigned to memory. |
AIVORY_LLM_OFFICIAL_TOOL_SPEC | string | "medium" | Transfer the search_context_size value to the OpenAI official built-in web_search tool specification. |
AIVORY_LLM_MAX_ITER_2 | int | 20 | In the OpenAI streamChat (Chat Completions) cycle, the native tool calls the rounds of the hardness upper limit. |
AIVORY_LLM_MAX_ITER_3 | int | 20 | The hardness upper limit of the native tool call rounds in the OpenAI streamResponses (Responses API) cycle. |
AIVORY_LLM_INLINE_QUOTE_SOURCE_INJECTION_CAP | int | 8000 | In the inbound reference sub conversation, the maximum number of characters (rune) in the source message text before the intersection is injected together with the high-bright excerpt. |
AIVORY_LLM_SANDBOX_EXEC_TIMEOUT_CLAMP_RANGE_MAX | int | 600 | When the context counting time for python_execute is called, the administrator-configured sandbox_exec_timeout_sec is limited to the upper limit (sec). |
AIVORY_LLM_SANDBOX_EXEC_TIMEOUT_CLAMP_RANGE_MIN | int | 10 | When the context counting time for python_execute is called, the administrator-configured sandbox_exec_timeout_sec is limited to the bottom limit (sec). |
AIVORY_LLM_SANDBOX_EXEC_CTX_SAFETY_MARGIN | duration | 150*time.Second | When counting the context time for python_execute, the additional surplus is added over the sandbox executing overtime, making the context later overtime than the sandbox HTTP client. |
AIVORY_LLM_PER_TURN_TOOL_LIMITS_WEB_SEARCH | int | 16 | The maximum number of web_search calls allowed for each message in the normal mode is exceeded; this means that the call fails. |
AIVORY_LLM_PER_TURN_TOOL_LIMITS_WEB_FETCH | int | 12 | The maximum number of web_fetch calls allowed for each message in the normal mode; exceeding that means that the call fails. |
AIVORY_LLM_PER_TURN_TOOL_LIMITS_FETCH_IMAGE | int | 16 | The maximum number of fetch_image calls allowed for each message in the normal mode; exceeding that means that the call fails. |
AIVORY_LLM_PER_TURN_TOOL_LIMITS_IMAGE_GENERATE | int | 8 | The maximum number of image_generate calls allowed for each message in the normal mode is exceeded; this means that the call fails. |
AIVORY_LLM_PER_TURN_TOOL_LIMITS_PYTHON_EXECUTE | int | 16 | The maximum number of times that a python_execute sandbox can execute for each message is allowed in the normal mode; exceeding that means that the call fails. |
AIVORY_LLM_DEEP_RESEARCH_TOOL_LIMITS_WEB_SEARCH | int | 40 | The maximum number of web_search calls allowed for each message when Deep Research is running is exceeded, which causes the call to fail. |
AIVORY_LLM_DEEP_RESEARCH_TOOL_LIMITS_WEB_FETCH | int | 25 | The maximum number of web_fetch calls allowed for each message when Deep Research is running is exceeded, which causes the call to fail. |
AIVORY_LLM_DEEP_RESEARCH_TOOL_LIMITS_FETCH_IMAGE | int | 12 | The maximum number of fetch_image calls allowed for each message when Deep Research is running is exceeded; this means that the call fails. |
AIVORY_LLM_DEEP_RESEARCH_TOOL_LIMITS_IMAGE_GENERATE | int | 4 | The maximum number of image_generate calls allowed for each message when Deep Research is running is exceeded, which causes the call to fail. |
AIVORY_LLM_DEEP_RESEARCH_TOOL_LIMITS_PYTHON_EXECUTE | int | 8 | The maximum number of times a python_execute sandbox is allowed to execute each message while Deep Research runs; exceeding that means that the call fails. |
AIVORY_LLM_MAX_TOOL_CALLS_PER_TURN | int | 48 | In the ordinary mode, each message calls the total global ceiling of the total number of tools across all tools, overlapping over the separate ceiling of each tool. |
AIVORY_LLM_MAX_TOOL_CALLS_PER_TURN_DEEP | int | 150 | Deep Research runs a global ceiling on the total number of calls for each message across all tools. |
AIVORY_LLM_TOOL_TIMEOUTS | duration | 10*time.Second | A single web_search tool calls a time limit for each call. |
AIVORY_LLM_TOOL_TIMEOUTS_2 | duration | 15*time.Second | A single web_fetch tool calls a time limit for each call. |
AIVORY_LLM_TOOL_TIMEOUTS_3 | duration | 600*time.Second | The image_generate tool calls a single time limit for each call (leaving a wider window for delayed third-party image gateways). |
AIVORY_LLM_TOOL_TIMEOUT_DEFAULT | duration | 100*time.Second | Return time for every call used by the tool listed in the toolTimeouts mapping that is not type-configured. |
AIVORY_LLM_PROMPT_MAX_ITER | int | 10 | The maximum number of iterations before the end of the round (each iteration = one model generated plus an optional tool call). |
AIVORY_LLM_PROMPT_MAX_RETRY | int | 2 | Repeat the error format. <tool_call> The maximum number of repeat attempts performed by the JSON or failure tool. |
AIVORY_LLM_PROVIDER_REQUEST_BODY_MAX_BYTES | int | 128*1024 | The maximum length of the captured supplier request body/request head JSON snapshot used for diagnosis is captured. |
AIVORY_LLM_PROVIDER_REQUEST_VALUE_MAX_BYTES | int | 8*1024 | Each individual captured supplier request value (URL, request head value) is the maximum length that is captured. |
AIVORY_LLM_IMAGE_DOCUMENT_FLAT_TOKEN_ALLOWANCE | int | 1024 | Estimate the fixed number of tokens counted for each image/document block when requesting tokens, because base64 cannot be divided by text. |
AIVORY_LLM_OUTPUT_RESERVE | int | 2000 | When calculating the predictive credits affordability estimate, estimate the amount of fixed output token reserves added above the input token. |
AIVORY_LLM_MAX_CONCURRENT_TOOLS | int | 4 | Maximum number of tools executed simultaneously in a single wheel (limiting the signal capacity of the simultaneous executor). |
AIVORY_LLM_VCTX | duration | 45*time.Second | Limit overtime calls for post-response verification/audit LLM to avoid delayed cross-supplier audits. |
RAG Document Analysis/Vector Recovery
Document splits, Embedding batches and parallels, Qdrant clients, MinerU OCR rounds and more.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_RAG_EMBEDDING_RETRY_DELAY | duration | time.Second | The base delay of index evasion between embedded batches is amplified by 2^min (number of attempts, 4). |
AIVORY_RAG_EMBEDDING_RETRY_DELAY_2 | duration | 30*time.Second | The upper limit for the delay being avoided is calculated for embedded re-trial. |
AIVORY_RAG_EMBEDDING_RETRY_DELAY_3 | duration | 1000*time.Millisecond | The random shaking range added for each embedding re-trial to avoid delay is used for the synchronization of scattered and batch-outs. |
AIVORY_RAG_DASH_SCOPE_EMBED_CONCURRENCY | int | 2 | When the provider is DashScope, the upstream embedded in each document parallels the upper limit of the batch. |
AIVORY_RAG_DASH_SCOPE_GLOBAL_EMBED_CONCURRENCY | int | 2 | Process-level parallel DashScope embeds the upper limit on the number of requests (capacity of shared slot signals). |
AIVORY_RAG_DASH_SCOPE_EMBED_ATTEMPT_TIMEOUT | duration | 60*time.Second | DashScope embedded a single attempt call requests overtime, set below the client's upper 3 minute limit, making stuck calls visibly fail. |
AIVORY_RAG_EMBED_CONCURRENCY | int | 4 | A non-DashScope provider limits the upstream batch of parallel embedding for each document. |
AIVORY_RAG_MAX_ATTEMPTS | int | 2 | A single embedded batch of POST represents the maximum number of attempts before giving up (removing attempts in case of instant failure). |
AIVORY_RAG_PDF_INSPECTION_TIMEOUT | duration | 8*time.Second | The parent process pair of subprocesses PDF text layer detection probes impose a deadline that is about to kill them. |
AIVORY_RAG_OFFICE_XML_ZIP_ENTRY_READ_CAP | int64 | 16*1024*1024 | When Office Pure Text is extracted, read the upper limit of bytes for each DOCX/PPTX zip entry (text/headline/foot/XML slide). |
AIVORY_RAG_PDF_INSPECTION_SAMPLE_LIMIT | int | 3 | Detect the maximum number of pages that are uniformly sampled when a PDF is a scan or native digital text. |
AIVORY_RAG_PDF_THIN_CHARS_PER_PAGE | int | 200 | When the threshold for the number of characters per sample page is lower than this, the graphic PDF is marked as "text thin" and transmitted to OCR. |
AIVORY_RAG_CMD_WAIT_DELAY | duration | 500*time.Millisecond | After the PDF detector subprocesses overtime, the mandatory pre-killing grace period (cmd.WaitDelay) is applied. |
AIVORY_RAG_PDF_RAW_SIGNALS_STRONG_SCAN_IMAGE | int | 5 | The original byte “Strong Scanner” determines the image multiplier in imageCountN >= pagesM, which is used to determine that the PDF is a pure image. |
AIVORY_RAG_PDF_RAW_SIGNALS_STRONG_SCAN_PAGE | int | 4 | The original byte “Strong Scanner” determines the page multiplier in imageCountN >= pagesM, which is used to determine that the PDF is a pure image. |
AIVORY_RAG_MINERU_SOURCE_OBJECT_CLEANUP_TIMEOUT | duration | 30*time.Second | After the analysis, try to remove the time limit for the inbound objects uploaded to MinerU OCR. |
AIVORY_RAG_MINERU_POLL_DEADLINE | duration | 20*time.Minute | The rotating MinerU extract task waits for the total time limit of the entire cycle for the done/failed state. |
AIVORY_RAG_MINERUZIPCLIENT_TIMEOUT | duration | 5*time.Minute | HTTP client with SSRF protection downloads MinerU result zip. |
AIVORY_RAG_FULL_MD_READ_CAP_INSIDE_ZIP | int64 | 32*1024*1024 | Read the upper limit of the byte of the full.md markdown text from the MinerU result zip. |
AIVORY_RAG_MAX_ZIP | int64 | 500*1024*1024 | Download the upper byte limit for the MinerU result zip; exceeding is rejected without buffering into memory. |
AIVORY_RAG_MINERUCLIENT_TIMEOUT | duration | 5*time.Minute | Call the MinerU Cloud API to submit, rotating queries, and downloading the http.Client overtime each time. |
AIVORY_RAG_MINERU_SOURCE_TTLSECONDS | int | 60*60 | The pre-signature GET URL for the source document delivered to MinerU is valid (seconds). |
AIVORY_RAG_RAG_FAST_QUEUE_CONCURRENCY | int | 4 | The asynq work parallel number of the "rag-fast" intake channel for the service text/table document. |
AIVORY_RAG_RAG_SLOW_QUEUE_CONCURRENCY | int | 4 | Service OCR/heavy document "rag" asynq work parallel numbers for slow intake channels. |
AIVORY_RAG_INGEST_PIPELINE_TIMEOUT | duration | 70*time.Minute | The full analysis/embedded timing of each document running runIngestWithRetries. |
AIVORY_RAG_INGEST_TASK_TIMEOUT | duration | 75*time.Minute | Asynq is a single task processing overtime (asynq.Timeout) for rag.ingest tasks. |
AIVORY_RAG_INGEST_UNIQUE_TTL | duration | 80*time.Minute | Asynq is a unique sex lock TTL (asynq.Unique), which inhibits repeated entry of the same document. |
AIVORY_RAG_INGEST_HEARTBEAT_INTERVAL | duration | 30*time.Second | Capture the intervals of TouchDocumentIngest so that the running document is not deemed to be expired. |
AIVORY_RAG_INGEST_STALE_AFTER | duration | 4*time.Minute | The heartbeat of the document that is analyzed/embedded for more than this time is considered to be abandoned and recovered to re-enter. |
AIVORY_RAG_INGEST_PENDING_STALE_AFTER | duration | ingestUniqueTTL | Documents that are still in "pending" for more than this period of time are considered to be stuck and recovered to re-enter. |
AIVORY_RAG_INGEST_RECOVERY_INTERVAL | duration | time.Minute | Recover the scan interval for the recovery cycle of expired abandoned intake documents. |
AIVORY_RAG_INGEST_FINALIZE_TIMEOUT | duration | 30*time.Second | At the end of the failed intake, the vector library DeleteByDocument clears the overtime. |
AIVORY_RAG_INGEST_ASYNQ_LEASE_MAX_RETRIES | int | 1 | The number of times asynq MaxRetry is lost reserved for the lease/process; the processor fails to try again on its own in the streamline. |
AIVORY_RAG_INGEST_ASYNQ_RETRY_DELAY | duration | 2*time.Minute | asynq restarts the waiting delay (RetryDelayFunc) before taking the missed rental task. |
AIVORY_RAG_INGEST_QUEUE_NAME | duration | 2*time.Second | GetDocument query overtime for dividing documents into fast/slow take cohorts. |
AIVORY_RAG_RUN_INGEST_WITH_RETRIES | int | 3 | The document was terminated as failed before the maximum number of attempts to ingest the entire streamline. |
AIVORY_RAG_RUN_INGEST_WITH_RETRIES_2 | duration | 3*time.Second | The entire flow water line intake base evasion between re-tests, multiplied by the number of attempts. |
AIVORY_RAG_START_INGEST_HEARTBEAT | duration | 5*time.Second | The time limit for each intake of the heartbeat is written (TouchDocumentIngest). |
AIVORY_RAG_FINALIZE_CHUNK_CLEANUP_TIMEOUT | duration | 10*time.Second | At the end of the failed intake, DeleteChunksByDocument clears the overtime. |
AIVORY_RAG_FINALIZE_STATUS_TIMEOUT | duration | 10*time.Second | Mark the failed document as “failed” in the final UpdateDocumentStatus overtime. |
AIVORY_RAG_DENSE_SEARCH_LEG_LIMIT | int | 30 | The number of Top-N commands requested by the density/vec.Search branch in the hybrid search. |
AIVORY_RAG_KEYWORD_SEARCH_LEG_LIMIT | int | 30 | The number of Top-N commands requested by the search keyword branch (vec.SearchKeyword) in the hybrid search. |
AIVORY_RAG_SNIPPET_OF | int | 240 | When the caller does not specify a ceiling, the result summary has the default byte length ceiling (by rune secure intersection). |
AIVORY_RAG_SPLIT_PARAGRAPHS_AND_TABLES | int | 800 | Picture markdown paragraphs below this byte length are retained as single atomic blocks and no longer divided. |
AIVORY_RAG_ROUTER_CALL_TIMEOUT | duration | 12*time.Second | The first token hot path routes tasks to the JSON LLM call (task.router) over time before returning to normal retrieval. |
AIVORY_RAG_MAP_REDUCE_SUMMARISE | int | 200 | The upper limit (≤N word) for the number of middle words requested to the task model in the map-reduce segment summary tip. |
AIVORY_RAG_COLLECT_DOC_HINTS | int | 120 | The router document reminder line preserves the number of lead bytes for each document's original text to remove the reminder. |
AIVORY_RAG_COLLECT_DOC_HINTS_2 | int | 12 | Maximum number of prompts per document collected for retrieval by the router. |
AIVORY_RAG_FUSE_RECIPROCAL_RANK | int | 60 | The fusion vector and keyword retrieval reverse ranking constants in the fusion formula 1/(rank+k). |
AIVORY_RAG_RETRIEVED_SNIPPET_CHARS | int | 2000 | expandHit injects the byte budget built around each pick-up command block before injecting. |
AIVORY_RAG_CHILD_TARGET_CHARS | int | 2000 | The target byte length that tends to each embedded block when cutting the atomic unit. |
AIVORY_RAG_PARENT_TARGET_CHARS | int | 4800 | The parent-level chapter is the byte length that was cut from the text when it was retrieved. |
AIVORY_RAG_CHUNK_OVERLAP_CHARS | int | 250 | As a sliding window overlap, add the number of bytes of the tail of the first block in front of the next subblock. |
AIVORY_RAG_MAPREDUCE_GROUPTOKENS | int | 6000 | Make a map-reduce summary of the estimated token budget for each block. |
AIVORY_RAG_MAPREDUCE_MAXGROUPS | int | 8 | The map-reduce summary group the maximum number of blocks to be processed for a single query. |
AIVORY_RAG_BATCH_SIZE | int | 64 | Manages the number of embedded blocks each time the Embed calls when reconstructing the vector. |
AIVORY_VECTOR_QDRANT_SCROLL_PAGE_SIZE_EXISTINGCHUNKIDS | int | 256 | List the number of points obtained by each Qdrant scroll page when the block ID already exists in the index. |
AIVORY_VECTOR_QDRANT_SCROLL_PAGE_SIZE_VECTORCHUNKSTATUSES | int | 256 | To manage the end vector status audit scan load and the number of points obtained for each Qdrant scroll page when vectoring. |
AIVORY_VECTOR_DELETE_CONCURRENCY | int | 4 | The maximum number of deletion requests subsequently sent by collection when a document is cleared across all dimensions. |
Execution of sandbox code
python_execute Go side variables need to restart the API process to take effect; SANDBOX_* The variable works. sandbox-service The process must resume the service.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
SANDBOX_MAX_OUTPUT_BYTES | int | 32768 | The sandbox cuts the upper byte limit of the exec stdout/stderr stream captured. |
SANDBOX_MAX_ARTIFACT_BYTES | int | 20971520 | Sandbox collects the maximum byte size of a single output file. |
SANDBOX_S3_MAX_ATTEMPTS | int | 3 | Limit the number of repetitions of botocore max_attempts for each S3 storage SDK call. |
SANDBOX_S3_CONNECT_TIMEOUT_S | float | 10 | Connection overtime (seconds) applied to each S3 storage SDK call. |
SANDBOX_S3_READ_TIMEOUT_S | float | 120 | Apply reading overtime (seconds) to each S3 storage SDK call. |
SANDBOX_OSS_CONNECT_TIMEOUT_S | float | 30 | OSS bucket client connectivity overtime (seconds) |
AIVORY_SANDBOX_MAX_SANDBOX_RESP_BYTES | int64 | 256<<20 | Each exec call decodes the sidecar HTTP responsive byte upper limit to prevent API process OOM. |
AIVORY_SANDBOX_EXEC_CLIENT_OVERHEAD | duration | 120*time.Second | The length of time added above each exec upper limit is used to set the sandbox HTTP client overtime to cover the sidecar deadline. |
AIVORY_SANDBOX_SANDBOX_ERROR_BODY_READ_CAP | int64 | 64<<10 | Read the 4xx/5xx sidecar error respondent for the upper byte of the error information. |
Built-in tools (Search / Python / Cyber Security)
Web_search results with overtime, Python Security Mode, SSRF/Network Security Bar and more.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_TOOLS_IN_TOP_K | int | 5 | model calls the default number of results for the web_search tool when top_k is not specified. |
AIVORY_TOOLS_WEB_FETCH_RESPONSE_BODY_READ_CAP | int64 | 256*1024 | web_fetch reads the upper limit of the HTTP-responsive byte before decomposing HTML. |
AIVORY_TOOLS_WEB_FETCH_EXTRACTED_TEXT_CHAR_CAP | int | 32000 | web_fetch marks the upper limit of characters before the intersection to omit the text after the HTML is removed. |
AIVORY_TOOLS_PYTHON_EXECUTE_UPLOAD_STAGING_FILE_SIZE | int64 | 20*1024*1024 | When a file is temporarily stored in a python sandbox, the value is exceeded by skipping the upper limit of the single file size for a session upload. |
AIVORY_TOOLS_PYTHON_EXECUTE_IMAGE_ARTIFACT_STAGING_SIZE | int64 | 20*1024*1024 | When temporarily stored in the python sandbox, the value exceeds the upper limit for the single file size of a generated image product. |
AIVORY_TOOLS_PYTHON_EXECUTE_STDOUT_STDERR_TRUNCATION_CAP | int | 32*1024 | python_execute's stdout/stderr is the upper limit of the character cut before presenting to the model. |
AIVORY_TOOLS_IN_N | int | 4 | A single image_generate calls the maximum number of images that can be requested. |
AIVORY_TOOLS_IN_SIZE | string | "1024x1024" | image_generate calls the default width × high image size used when the size is not specified. |
AIVORY_TOOLS_DAILY_IMAGE_LIMIT_RESET_WINDOW | duration | 24*time.Hour | Calculate the time window for Now() when generating the quota book starting boundary for each user image. |
AIVORY_TOOLS_IMAGE_IMAGE_INPUT_IMAGE_CAP | int | 3 | The diagram calls the maximum number of loaded reference input images. |
AIVORY_TOOLS_FETCHREMOTEIMAGE_DOWNLOAD_CAP | int64 | 32<<20 | Download the upper limit of bytes for image URLs returned in image API responses via the SSRF Secure Client. |
AIVORY_TOOLS_IN_TOP_K_2 | int | 5 | The default number of fragments for the search_knowledge_base tool is called when top_k is not specified. |
AIVORY_TOOLS_CONFIDENCE | float | 0.95 | The confidence score is stored on each memory record created by the save_memory tool. |
AIVORY_TOOLS_MAX_IMG | int64 | 15*1024*1024 | The fetch_image tool downloads the upper limit of the byte of the image; beyond rejects the response. |
Session / Message / Streaming API
SSE heartbeat, flow recovery window, generate time upper limit, page and search upper limit, message path cache and more.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_API_LIMIT_2 | int | 200 | The default subpage size of the list-conversations interface when the limit argument is not provided for the request. |
AIVORY_API_LIMIT_3 | int | 500 | Session list interface ?limit The hardness upper limit of the query parameter exceeds that value. |
AIVORY_API_SEARCH_MESSAGE_HIT_LIMIT | int | 40 | The upper limit of the number of messages returned by the session search interface (there is a separate upper limit in the title command). |
AIVORY_API_IMPORT_MAX_CONVERSATIONS | int | 1000 | The upper limit for the number of sessions that can be created for a single import request is refused to import. |
AIVORY_API_IMPORT_MAX_MESSAGES_PER_CONV | int | 10000 | Import the maximum number of messages allowed for a single session in the request and refuses to import beyond. |
AIVORY_API_IMPORT_MAX_CONTENT_BYTES | int | 200*1024 | When importing a single line message text, the byte upper limit exceeds the section cut by this byte number. |
AIVORY_API_INLINE_THREAD_QUOTE_CAP | int | 4000 | The maximum rune length of the text is referenced when creating an intra-line sub-theme, exceeding the intersection. |
AIVORY_API_GETCONVERSATION_ACTIVE_PATH_LIMIT | int | 200 | Activity path pages are divided by the upper limit of the limit parameter when a session is obtained; the excess returns the full path. |
AIVORY_API_LIMIT_4 | int | 30 | The message list interface does not provide the default tip window page size when limit. |
AIVORY_API_LISTMESSAGES_PAGE_LIMIT | int | 200 | The message list interface is the maximum acceptable value for the limit parameter; above is the default page size. |
AIVORY_API_RATE_LIMIT_USER | int | 20 | Traffic limit per user: The maximum number of times a knowledge base document is allowed to be uploaded per minute exceeds 429 returns. |
AIVORY_API_CONFIDENCE | float | 0.95 | The confidence score that is stored in the user's manually created memory entry (0-1 interval). |
AIVORY_API_MAX_GEN_DURATION | duration | 90*time.Minute | Single-separation generated rounds of the clock overtime ceiling, after overtime cancelled its content. |
AIVORY_API_SSE_PING_HEARTBEAT_POST | duration | 15*time.Second | Send message streaming interfaces with SSE guaranteed ping sending intervals. |
AIVORY_API_SSE_PING_HEARTBEAT_REGENERATE | duration | 15*time.Second | Re-generate the sending intervals of the SSE guaranteed ping on the message stream interface. |
AIVORY_API_SSE_PING_HEARTBEAT_STREAM | duration | 15*time.Second | SSE preserves the sending interval for ping on the attachment interface. |
AIVORY_API_STREAM_STATUS_RECHECK_INTERVAL | duration | 5*time.Second | The flow re-connector processor rotates the message-generated state to detect the end interval. |
AIVORY_API_STREAM_REPLAY_BATCH_SIZE | int | 200 | Repeat the number of buffering events read per batch when SSE flow is returned/tracked. |
AIVORY_API_ONLINE_PRESENCE_TOUCH_THROTTLE | duration | time.Minute | The user’s online state seen touches between the intervals, i.e. the TTL of the seen cache marker. |
AIVORY_API_CONCURRENT_GEN_SLOT_SAFETY_TTL | duration | 30*time.Minute | Each user and the safe TTL of the slot counter occurs, so that the invalid occupied slot automatically expires. |
AIVORY_API_REQUEST_SIGNATURE_REPLAY_WINDOW_FUTURE | int64 | 300 | Request Signature Reset: X-Req-Ts time stamps the maximum number of seconds of backward server time, exceeding which is considered to be rejected. |
AIVORY_API_REQUEST_SIGNATURE_REPLAY_WINDOW_PAST | int64 | 60 | Request Signature Reset: The X-Req-Ts time stamp leads the maximum number of seconds of server time (clock shift tolerance) and rejects the excess. |
AIVORY_API_CREDIT_MULTIPLIER | float | 5.0 | Convert the model input + output total price to the deduction of the relative credits multiplier in the selector. |
AIVORY_API_JSON_REQUEST_BODY_SIZE_CAP | int64 | 4<<20 | The maximum number of bytes acceptable for JSON requests exceeds the number rejected by MaxBytesReader. |
AIVORY_API_PROJECT_DETAIL_CONVERSATIONS_PAGE_SIZE | int | 200 | Number of activity sessions loaded in the project details response (acquired upper limit, shifted to 0). |
AIVORY_API_RATE_LIMIT_USER_2 | int | 20 | Traffic limit per user: The maximum number of times that project documents are allowed to be uploaded per minute exceeds 429 returns. |
AIVORY_API_RATE_LIMIT_REGISTER_MAX | int | 5 | POST/api/auth/register Traffic Limit Per IP: The maximum number of registration attempts allowed in each window, exceeding the return of 429. |
AIVORY_API_RATE_LIMIT_REGISTER_WINDOW | duration | 60*time.Second | POST /api/auth/register The length of the scroll window for the flow limit per IP. |
AIVORY_API_RATE_LIMIT_LOGIN_MAX | int | 10 | Traffic limit per IP for POST/api/auth/login: The maximum number of log-in attempts permitted in each window, exceeding the return of 429. |
AIVORY_API_RATE_LIMIT_LOGIN_WINDOW | duration | 60*time.Second | POST /api/auth/login The length of the scroll window for the flow limit per IP. |
AIVORY_API_RATE_LIMIT_LOGIN_2FA_MAX | int | 10 | POST /api/auth/login/2fa Traffic Limits per IP: The maximum number of two-step authentication attempts allowed in each window exceeds the return of 429. |
AIVORY_API_RATE_LIMIT_LOGIN_2FA_WINDOW | duration | 60*time.Second | POST /api/auth/login/2fa The length of the scroll window for the flow limit per IP. |
AIVORY_API_RATE_LIMIT_LOGOUT_MAX | int | 30 | POST/api/auth/logout Traffic Limits per IP: The maximum number of outbound requests allowed in each window, exceeding returns of 429. |
AIVORY_API_RATE_LIMIT_LOGOUT_WINDOW | duration | 60*time.Second | POST /api/auth/logout The length of the scroll window for the flow limit per IP. |
AIVORY_API_RATE_LIMIT_REFRESH_MAX | int | 30 | POST/api/auth/refresh Traffic Limit Per IP: The maximum number of token refresh requests allowed in each window exceeds 429 returns. |
AIVORY_API_RATE_LIMIT_REFRESH_WINDOW | duration | 60*time.Second | POST /api/auth/refresh The length of the scroll window for the flow limit per IP. |
AIVORY_API_RATE_LIMIT_VERIFY_EMAIL_MAX | int | 10 | POST/api/auth/verify-email Traffic Limit Per IP: The maximum number of mailbox authentication attempts allowed in each window exceeds 429 returns. |
AIVORY_API_RATE_LIMIT_VERIFY_EMAIL_WINDOW | duration | 5*60*time.Second | POST /api/auth/verify-email The length of the scroll window for the flow limit per IP. |
AIVORY_API_RATE_LIMIT_SEND_CODE_MAX | int | 3 | Each IP represents the maximum number of requests to the POST /api/auth/send-code interface in the window. |
AIVORY_API_RATE_LIMIT_SEND_CODE_WINDOW | duration | 60*time.Second | A fixed time window for statistics of the number of requests for /api/auth/send-code interfaces per IP. |
AIVORY_API_RATE_LIMIT_FORGOT_PASSWORD_MAX | int | 5 | Each IP represents the maximum number of requests to the POST/api/auth/forgot-password interface in the window. |
AIVORY_API_RATE_LIMIT_FORGOT_PASSWORD_WINDOW | duration | 15*60*time.Second | A fixed time window for statistics of the number of /api/auth/forgot-password interface requests per IP. |
AIVORY_API_RATE_LIMIT_RESET_PASSWORD_MAX | int | 5 | Each IP is the maximum number of requests for the POST/api/auth/reset-password interface in the window. |
AIVORY_API_RATE_LIMIT_RESET_PASSWORD_WINDOW | duration | 60*time.Second | A fixed time window for statistics of the number of /api/auth/reset-password interface requests per IP. |
AIVORY_API_RATE_LIMIT_CAPTCHA_ISSUE_MAX | int | 30 | Each IP represents the maximum number of requests for the GET/api/public/captcha interface in the window. |
AIVORY_API_RATE_LIMIT_CAPTCHA_ISSUE_WINDOW | duration | 60*time.Second | Statistics of the number of interface requests issued for the GET /api/public/captcha challenge for each IP. |
AIVORY_API_RATE_LIMIT_CAPTCHA_VERIFY_MAX | int | 60 | Each IP represents the maximum number of requests to the POST/api/public/captcha/verify interface in the window. |
AIVORY_API_RATE_LIMIT_CAPTCHA_VERIFY_WINDOW | duration | 60*time.Second | A fixed time window statistics the number of /api/public/captcha/verify response verification interface requests per IP. |
AIVORY_API_RATE_LIMIT_FIRST_RUN_SETUP_MAX | int | 10 | Each IP represents the maximum number of requests for the POST/api/setup (first run to create the first administrator) interface in the window. |
AIVORY_API_RATE_LIMIT_FIRST_RUN_SETUP_WINDOW | duration | 60*time.Second | Statistics of the fixed time window for each IP for the number of boot interface requests that POST/api/setup runs for the first time. |
AIVORY_API_RATE_LIMIT_PUBLIC_SHARED_CONVERSATION_MAX | int | 60 | Each IP is the maximum number of requests for the GET /api/public/shared/:token interface in the window. |
AIVORY_API_RATE_LIMIT_PUBLIC_SHARED_CONVERSATION_WINDOW | duration | 60*time.Second | Statistics for each IP for /api/public/shared/:token share a fixed time window for viewing the number of interface requests. |
AIVORY_API_RATE_LIMIT_SHARED_ASSETS_FILES_ARTIFACTS_MAX | int | 240 | Each IP is the maximum number of requests for file and artifact resource routing for a shared session in the window. |
AIVORY_API_RATE_LIMIT_SHARED_ASSETS_FILES_ARTIFACTS_WINDOW | duration | 60*time.Second | A fixed time window for the number of requested routes for sharing session files/product resources per IP. |
AIVORY_API_RATE_LIMIT_OAUTH_START_CALLBACK_HANDOFF_MAX | int | 20 | The maximum number of requests for OAuth initiation, referral and cross-domain routing per IP in the window. |
AIVORY_API_RATE_LIMIT_OAUTH_START_CALLBACK_HANDOFF_WINDOW | duration | 60*time.Second | Statistics of the fixed time window for the number of OAuth initiated/recalled/delivered routing requests per IP. |
AIVORY_API_RATE_LIMIT_PASSWORD_CHANGE_SET_MAX | int | 5 | Each IP is the maximum number of requests for /api/me/password transcripting and /api/me/password/set interfaces in the window. |
AIVORY_API_RATE_LIMIT_PASSWORD_CHANGE_SET_WINDOW | duration | 60*time.Second | A fixed time window statistics the number of /api/me/password encryption/set interface requests per IP. |
AIVORY_API_RATE_LIMIT_IDENTITY_LINK_START_MAX | int | 20 | The maximum number of requests per IP to the POST /api/me/identities/:id/link interface in the window. |
AIVORY_API_RATE_LIMIT_IDENTITY_LINK_START_WINDOW | duration | 60*time.Second | A fixed time window for statistics of the number of interface requests to initiate an OAuth identity association per IP. |
AIVORY_API_RATE_LIMIT_2FA_SETUP_ENABLE_DISABLE_MAX | int | 10 | The maximum number of requests per IP for /api/me/2fa setup, enable, and disable interfaces in the window. |
AIVORY_API_RATE_LIMIT_2FA_SETUP_ENABLE_DISABLE_WINDOW | duration | 5*60*time.Second | A fixed time window statistics the number of 2FA setup/enable/disable interface requests per IP. |
AIVORY_API_RATE_LIMIT_REDEEM_CODE_MAX | int | 10 | The maximum number of requests per IP to the POST/api/me/redeem (redeem code exchange) interface in the window. |
AIVORY_API_RATE_LIMIT_REDEEM_CODE_WINDOW | duration | 60*time.Second | A fixed time window for statistics of the number of requests for /api/me/redeem exchange interfaces per IP. |
AIVORY_API_RATE_LIMIT_WORKSPACE_JOIN_MAX | int | 30 | Each IP represents the maximum number of requests for the /api/workspaces/join/:token interface in the window. |
AIVORY_API_RATE_LIMIT_WORKSPACE_JOIN_WINDOW | duration | 60*time.Second | Statistics of the number of /api/workspaces/join/:token interface requests per IP. |
AIVORY_API_SELF_USAGE_LOOKBACK_WINDOW | int | 30 | GET /api/me/usage The number of days that the caller’s own message dates back. |
AIVORY_API_LIMIT | int | 200 | The default maximum range of work spaces returned when the request does not contain ?limit. |
AIVORY_API_ADMIN_WORKSPACE_DETAIL_CONVERSATIONS_PAGE_SIZE | int | 500 | Manage the maximum number of sessions loaded in the triage view. |
AIVORY_QUEUE_IN_PROCESS_WORKERS | int | 8 | The number of simultaneous worker syntax in the background task pool within the process. |
AIVORY_QUEUE_PROCESS_JOB_BUFFER | int | 256 | The number of buffer slots for the task channel in the process; when the buffer is full, Enqueue transfers the new task to the evasion-return path execution. |
AIVORY_QUEUE_QUEUE_BACKPRESSURE_JOB_TIMEOUT | duration | 30*time.Minute | In the process, the task buffer is overtime when the task executed by the back-to-back process is avoided. |
AIVORY_QUEUE_QUEUE_WORKER_JOB_TIMEOUT | duration | 30*time.Minute | In the process, the cohort worker performs the context overtime imposed on each task. |
Certification / Session / Verification Code
Token cache, validity code validity and number of attempts, TOTP time window, OAuth state TTL, etc. (excluding conventional format constants, such as verification code bits).
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_API_MAX_CODE_ATTEMPTS | int | 5 | The number of erroneous attempts to verify/reset the authentication code for the mail is reached and the code is canceled. |
AIVORY_API_CODE_FAILURE_COUNTER_TTL | duration | 10*time.Minute | The survival time (TTL) of the counter for the number of erroneous attempts to verify/reset the code recorded in the mailbox. |
AIVORY_API_MINIMUM_PASSWORD_LENGTH | int | 8 | The minimum character length required for the user's selected account password when registering and modifying the password. |
AIVORY_API_EMAIL_VERIFICATION_CODE_TTL | duration | 10*time.Minute | The validity period of the email sent account mailbox verification code. |
AIVORY_API_PASSWORD_RESET_CODE_TTL | duration | 10*time.Minute | The password sent by the email restores the validity of the authentication code. |
AIVORY_API_CAP_TOL | float | 0.04 | When the slide verification code passes, the allowed error between the submitted slide ratio and the actual gap ratio. |
AIVORY_API_CAPTCHA_CHALLENGE_CACHE_TTL | duration | 5*time.Minute | The unanswered scroll puzzle verification code challenge remains valid in the cache for as long as it remains valid. |
AIVORY_API_CAPTCHA_PASS_TTL | duration | 10*time.Minute | Prove the validity of the signature passed token that has recently passed the validation code. |
AIVORY_API_OAUTH_2FA_HANDOFF_COOKIE_TTL | duration | 300*time.Second | After OAuth logs in, the two-step authentication login ticket is sent to the front end of the short-term HttpOnly cookie Max-Age. |
AIVORY_API_OAUTH_STATE_CACHE_TTL | duration | 10*time.Minute | The OAuth process authorization state cache entry as CSRF protection is valid. |
AIVORY_API_OAUTH_TOKEN_EXCHANGE_CONTEXT_TIMEOUT | duration | 20*time.Second | OAuth recalls authorization codes in exchange for tokens and the upper time limit for pulling user information. |
AIVORY_API_OAUTH_CROSS_DOMAIN_HANDOFF_TOKEN_TTL | duration | 60*time.Second | The front end is used in exchange for the one-time cross-domain OAuth delivery token validity. |
AIVORY_API_2FA_LOGIN_TICKET_BURN_THRESHOLD | int64 | 5 | The number of error TOTP verification codes permitted for a two-step verification of the sign-in ticket reaches the number of annulments of the ticket. |
AIVORY_API_ISSUE_TWOFA_TICKET | duration | 5*time.Minute | Once the password is correct, the validity period of the two-step verification sign-in ticket issued during the TOTP authentication code must be entered. |
AIVORY_OAUTH_APPLE_CLIENT_SECRET_JWT_EXPIRY | duration | 30*time.Minute | The validity of the generated Apple OAuth client API key JWT. |
Upload / Documentation / Sharing
Image processing, storage cleaning cycles, direct split, sharing tokens, downloading cache TTLs and so on.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_API_ADMIN_ICON_UPLOAD_SIZE | int64 | 256*1024 | The maximum number of bytes accepted for the model icon image uploaded by the administrator. |
AIVORY_API_AUDIO_TRANSCRIPTION_UPSTREAM_HTTP_TIMEOUT | duration | 120*time.Second | Overtime for the HTTP request to forward audio to the upstream transcript API. |
AIVORY_API_AUDIO_TRANSCRIPTION_USER_RATE_LIMIT | int | 20 | The maximum number of audio transcript requests that a single user can initiate per minute. |
AIVORY_API_TRANSCRIPTION_UPSTREAM_RESPONSE_READ_CAP | int64 | 1<<20 | Read upstream transcripts responding to the upper limit of bytes in memory. |
AIVORY_API_UPLOAD_RATE_LIMIT_MAX | int | 20 | The maximum number of file uploads that a single user can carry out within the Upload Stream window. |
AIVORY_API_UPLOAD_RATE_LIMIT_WINDOW | duration | time.Minute | A window for limiting the flow counting of user file uploads. |
AIVORY_API_OBJECT_STORAGE_DELETE_TIMEOUT_CLEANUP | duration | 30*time.Second | Storage cleaning removes overtime for individual isolated files from object storage. |
AIVORY_STORAGE_S3_DIRECT_UPLOAD_MIN_CLIENT_TIMEOUT | duration | 20*time.Minute | The minimum HTTP client overtime required for direct S3/OSS upload; a new client is created when the reuse is lower than this value. |
AIVORY_STORAGE_DIRECT_S3_OSS_UPLOAD_HTTP_CLIENT | duration | 20*time.Minute | Repeat client overtime short, overtime for direct S3/OSS upload of newly built HTTP client. |
AIVORY_STORAGE_ALIYUN_OSS_CLIENT_CONNECT_READ_TIMEOUTS_CONNECT | int64 | 30 | Direct connection overtime (seconds) for the Ali Cloud OSS client used. |
AIVORY_STORAGE_ALIYUN_OSS_CLIENT_CONNECT_READ_TIMEOUTS_RW | int64 | 300 | Direct read time overwritten by the Ali Cloud OSS client (seconds). |
AIVORY_STORAGE_PRESIGN_URL_TTL | duration | 3600*time.Second | When the caller does not specify a validity date, the GET URL application pre-signature expires for direct transmission. |
AIVORY_STORAGE_PRESIGN_URL_TTL_CLAMP_CEILING | duration | 86400*time.Second | Improve the upper limit of the validity of the direct pre-signature GET URL requested by the caller. |
AIVORY_STORAGE_SIDECAR_STORAGE_CLIENT_HTTP_TIMEOUT | duration | 5*time.Minute | Call the sandbox sidecar object to store the put/delete interface for HTTP transit overtime. |
Admin console tasks (backup / vector maintenance / redeem code)
Backup size upper limit with asynchronous routing, vector maintenance batch, redeem code upper limit, etc.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_API_BACKUP_EXPORT_JOB_HISTORY_RETENTION | int | 20 | Previous backups retained in memory history export the number of task records, exceeding the number of oldest records after cutting. |
AIVORY_API_BACKUP_EXPORT_JOB_RUNTIME | duration | 12*time.Hour | Background backup exports take the maximum actual run time allowed before the task is cancelled. |
AIVORY_API_USER_DELETE_JOB_RUNTIME | duration | 2*time.Hour | The asynchronous user deletes the total run budget for the task, after the overtime task fails and waits for the restart to run (full-step scratch, etc.). |
AIVORY_API_USER_DELETE_FINAL_CLEANUP_TIMEOUT | duration | 1*time.Hour | Remove the standalone time budget for the task end disk/object storage cleaning phase (not occupying the main task budget); the value is also used to start the bottom scan. |
AIVORY_API_USER_DELETE_CONV_BATCH | int | 200 | Remove the number of sessions (along with its message) that are cleaned for each short transaction of the task, the smaller the batch, the more friendly the SQLite single writer is. |
AIVORY_API_USER_DELETE_USAGE_BATCH | int | 10000 | Remove the number of log lines that are used for cleaning each batch of tasks. |
AIVORY_API_USER_DELETE_JOB_HISTORY_RETENTION | int | 20 | Users stored in memory delete the number of tasks history records. |
AIVORY_API_USER_DELETE_PROGRESS_EVERY | int | 50 | The vector/disk cleaning phase shows how many items are updated for each process. |
AIVORY_API_CONFIG_IMPORT_MULTIPART_MEMORY_BUFFER | int64 | 16<<20 | Analysis administrators configure to import multipart uploaded memory buffers; exceed partial overwriting to temporary files. |
AIVORY_API_BACKUP_IMPORT_MULTIPART_MEMORY_BUFFER | int64 | 32<<20 | Resolve the full backup imported to the multipart uploaded memory buffer zone; beyond the partial overwriting to the temporary file. |
AIVORY_API_MAX_CONFIG_SIZE | int64 | 512<<20 | The maximum request size accepted when importing an administrator configuration archive. |
AIVORY_API_QDRANT_ARCHIVE_REQUEST_TIMEOUT | duration | 5*time.Minute | Export or import backup archives when calling Qdrant for a single HTTP request overtime. |
AIVORY_API_QDRANT_EXPORT_SCROLL_PAGE_SIZE | int | 256 | Export the number of credits drawn from each /points/scroll page when the Qdrant collection is archived. |
AIVORY_API_QDRANT_IMPORT_UPSERT_FLUSH_BATCH_SIZE | int | 128 | When importing a Qdrant collection, each batch accumulated to that score is printed upert once. |
AIVORY_API_ADMIN_USER_LIST_PAGE_SIZE_CAP | int | 50 | The administrator user interface returns the default number of bars when a limit is not explicitly specified. |
AIVORY_API_ADMIN_CREATED_USER_MIN_PASSWORD_LENGTH | int | 8 | The minimum password length required by the administrator when creating a user account. |
AIVORY_API_ADMIN_PASSWORD_RESET_MIN_LENGTH | int | 8 | The administrator resets the minimum character length required for the new password in the user password interface. |
AIVORY_API_ADMIN_USER_CONVERSATIONS_LISTING_CAP | int | 500 | Administrators view the maximum number of sessions returned for a single user session (for support/abuse scrutiny). |
AIVORY_API_USAGE_REPORT_PAGE_SIZE_CAP | int | 50 | The administrator records the default number of pages per page used in the report when the requested page_size is missing or exceeds the 1-200 range. |
AIVORY_API_ANALYTICS_WINDOW | int | 30 | The administrator analyzes the default reverse window (day) when the days query parameter is not provided. |
AIVORY_API_ANALYTICS_WINDOW_2 | int | 365 | Administrators analyze the days query parameters for the acceptable maximum number of days, and values beyond the range will be ignored. |
AIVORY_API_ANALYTICS_BREAKDOWN_TOP_N | int | 8 | The number of header keys retained (Top-N) in Administrator Analysis by model and by user segmentation and its time sequence. |
AIVORY_API_BULK_REDEEM_CODE_GENERATION_QUANTITY | int | 1000 | A single batch generates the maximum number of redeem codes that can be cast for a request. |
AIVORY_API_MAX_SKILL_ASSET_BYTES | int64 | 20*1024*1024 | Maximum number of bytes uploaded for a single skill resource file (template/script/small data). |
AIVORY_API_VECTOR_MAINTENANCE_JOB_HISTORY_RETENTION | int | 20 | The last number of tasks stored in the memory in the vector maintenance history will be discarded. |
AIVORY_API_VECTOR_MAINTENANCE_JOB_RUNTIME | duration | 12*time.Hour | A single vector maintenance task (index audit or reconstruction) runs overtime in the context before it is cancelled. |
Server Launch / Configuration
HTTP server overtime, elegant shutdown, start-up process constant and so on.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
AIVORY_CMD_ARCHIVE_GC_BOOT_SETTLE_DELAY | duration | 2*time.Minute | Delay between server boot and first archive work area GC cleaning, avoiding immediate cleaning when cold boot. |
AIVORY_CMD_RUN_PRUNE | duration | 5*time.Minute | A single archive GC cleaning workspace (for object storage) runs on the context overtime. |
AIVORY_CMD_ARCHIVE_GC_SWEEP_INTERVAL | duration | 6*time.Hour | The interval between the GC cleaning of the archive workspace is used to remove the expired /workspace archive package from the object storage. |
10 The front
** The translation period ** Entry into force (Vite in npm run build Internal time VITE_* The environmental variables need to be changed in the setting of the building environment and are invalid during the runtime. rotating intervals, page splitting, re-trial / evasion, shaking, overtime, client size upper limit, etc.
| Environmental variables | Type of | The default value | Explained |
|---|---|---|---|
VITE_AIVORY_IMAGE_API_MY_IMAGES_LIMIT | envNum | 60 | Draw the default number per page (limit) when the user is currently logged in to their own generated library. |
VITE_AIVORY_WORKSPACES_API_ADMIN_LIST_LIMIT | envNum | 200 | The default number (limit) per page requested by the Administrator’s Workspace list. |
VITE_AIVORY_CONVERSATIONS_API_LIST_LIMIT | envNum | 200 | The default number per page requested by the session list (limit). |
VITE_AIVORY_CONVERSATIONS_API_LIST_ARCHIVED_LIMIT | envNum | 200 | The default number (limit) per page requested by the archived session list. |
VITE_AIVORY_ADMIN_API_USERS_LIMIT | envNum | 50 | The administrator lists the default number per page request (limit). |
VITE_AIVORY_ADMIN_API_USER_IMAGES_LIMIT | envNum | 60 | The administrator scrolls down to see the default number per page (limit) when a user generates a chart. |
VITE_AIVORY_ADMIN_API_ANALYTICS | envNum | 30 | The administrator analyzes the default retrospect window (day) sent by the dashboard request. |
VITE_AIVORY_MAX_BYTES | envNum | 256 * 1024 | The administrator icon is the maximum number of bytes uploaded by the client, exceeding which is rejected before the request (in line with the backend upper limit). |
VITE_AIVORY_MAX_LEN | envNum | 12_000 | The maximum character length of the message is entered in the chat box, exceeding which is prohibited from sending. |
VITE_AIVORY_INGEST_POLL_MS | envNum | 1200 | During the time the sending button remains locked, the rotating appendix receives (resolves / embeds) the state interval (milliseconds). |
VITE_AIVORY_PAGE | envNum | 30 | The user generates an unlimited number of images scrolled per page (home screen loads and more loads). |
VITE_AIVORY_RUN_TIMEOUT_MS | envNum | 120_000 | Pyodide Python runs a single hard clock upper limit (milliseconds) in the browser, and overtime terminates worker. |
VITE_AIVORY_MAX_STREAM_CHARS | envNum | 200_000 | Pyodide runs the maximum total number of characters for midstream stdout/stderr, exceeding the cut to prevent out-of-control printing. |
VITE_AIVORY_MAX_RESULT_CHARS | envNum | 20_000 | Pyodide runs the final outcome expression repr() for the maximum character length before the intersection. |
VITE_AIVORY_ADMIN_BACKUP_EXPORT_JOB_POLL_INTERVAL | envNum | 2500 | The administrator backup page rotates to update the intervals (miles) for the status of the running backup export task. |
VITE_AIVORY_PAGE_SIZE_2 | envNum | 20 | Administrators redeem the number of rows per page in the code table (underpages for obtained rows in the client). |
VITE_AIVORY_PAGE_SIZE | envNum | 50 | The administrator records the number of rows per page requested by the table (pageSize). |
VITE_AIVORY_IMAGES_PAGE | envNum | 60 | Administrators scroll down the user's library to the number of images drawn per page in the view (home screen loads and more loads). |
VITE_AIVORY_ONLINE_WINDOW_S | envNum | 300 | In the Administrator User table, the last user activity falls within that time window (seconds) and is considered online. |
VITE_AIVORY_PAGE_SIZE_3 | envNum | 50 | The number of rows per page requested by the administrator's user list (service-side limit). |
VITE_AIVORY_KB_DOC_STATUS_POLL_INTERVAL | envNum | 2200 | When a document is still in the resolution/embedding phase, rotating refreshes the knowledge base document status intervals (milliseconds). |
VITE_AIVORY_CONV_PAGE | envNum | 200 | The side bar session list shows the number of sessions drawn per page (home screen loads and scroll loads). |
VITE_AIVORY_MSG_PAGE | envNum | 40 | The number of messages drawn per page when a session is opened (home page and the earlier message loaded by scrolling up). |