Environment Variables
All variables are optional. Constructor parameters take precedence over environment variables.
| Variable | Default | Description |
|---|---|---|
LLM_CHAT_WEB_SEARCH_PROVIDERS |
— | Comma-separated provider list, e.g. duckduckgo,tavily. Unset or empty resolves to a single keyless duckduckgo entry. Unknown names are skipped |
LLM_CHAT_WEB_SEARCH_PROVIDER_STRATEGY |
single |
How multiple providers are combined: single, parallel, fallback, weighted_random, or round_robin |
LLM_CHAT_WEB_SEARCH_TAVILY_API_KEY |
— | Per-provider API key for Tavily |
LLM_CHAT_WEB_SEARCH_EXAAI_API_KEY |
— | Per-provider API key for ExaAI |
LLM_CHAT_WEB_SEARCH_FIRECRAWL_API_KEY |
— | Per-provider API key for Firecrawl |
LLM_CHAT_WEB_SEARCH_MAX_RESULTS |
5 |
Default result count per search |
LLM_CHAT_WEB_SEARCH_MAX_CHARS_PER_RESULT |
2000 |
Max characters per result snippet |
LLM_CHAT_WEB_SEARCH_CONCURRENCY |
3 |
Concurrency limit for parallel searches |
LLM_CHAT_WEB_FETCH_CONCURRENCY |
3 |
Concurrency limit for parallel URL fetches |
LLM_CHAT_WEB_MAX_CHARS_PER_FETCH |
10000 |
Default max characters per URL fetch |
LLM_CHAT_WEB_MAX_CHARS_PER_FETCH_LIMIT |
100000 |
Hard limit for per-request max_chars override |
LLM_CHAT_WEB_FETCH_TIMEOUT_MS |
5000 |
Timeout in ms for URL fetch requests |
LLM_CHAT_WEB_SEARCH_TIMEOUT_MS |
4000 |
Timeout in ms for search API requests |
LLM_CHAT_WEB_MAX_CONTENT_LENGTH_BYTES |
10000000 |
Max bytes accepted before rejecting the fetch (checked via HEAD) |
LLM_CHAT_WEB_MAX_PDF_PAGES |
100 |
Default max pages extracted from a PDF (no hard cap; a per-call maxPages must not exceed it) |
WebSearchConfiguration
Constructor
An omitted or empty providers array triggers environment resolution (see Multi-provider mode). An explicit strategy overrides the environment value.
Properties
| Property | Type | Default | Env Variable | Description |
|---|---|---|---|---|
strategy |
ProviderStrategy |
Single |
LLM_CHAT_WEB_SEARCH_PROVIDER_STRATEGY |
How multiple providers are combined |
providers |
ProviderEntry[] |
resolved at construction | LLM_CHAT_WEB_SEARCH_PROVIDERS |
Active search backends (see below) |
maxResults |
number |
5 |
LLM_CHAT_WEB_SEARCH_MAX_RESULTS |
Default result count per search |
maxCharsPerResult |
number |
2000 |
LLM_CHAT_WEB_SEARCH_MAX_CHARS_PER_RESULT |
Default max chars per result snippet |
searchTimeoutMs |
number |
4000 |
LLM_CHAT_WEB_SEARCH_TIMEOUT_MS |
Timeout in ms for search API requests |
concurrency |
number |
3 |
LLM_CHAT_WEB_SEARCH_CONCURRENCY |
Concurrency limit for parallel searches |
Multi-provider mode
Each entry of the providers array is a ProviderEntry:
type ProviderEntry = {
provider: WebSearchProvider;
apiKey?: string; // Per-provider API key
weight?: number; // Selection weight used by weighted_random (default: 1)
};
Entries are resolved in this order (first match wins):
- Constructor argument (
providersarray), when non-empty. LLM_CHAT_WEB_SEARCH_PROVIDERSenv var: comma-separated provider names (e.g.duckduckgo,tavily). Unknown names are skipped. Each entry's API key comes from its dedicated per-provider env var.- A single keyless
duckduckgoentry, so zero-config setups keep working.
The list is never empty after construction. The strategy property selects how the entries are combined:
| Strategy | Env value | Behavior |
|---|---|---|
ProviderStrategy.Single |
single |
Query the first provider only |
ProviderStrategy.Parallel |
parallel |
Query all providers concurrently and merge results |
ProviderStrategy.Fallback |
fallback |
Try providers in order until one succeeds |
ProviderStrategy.WeightedRandom |
weighted_random |
Pick one provider per query via weight distribution |
ProviderStrategy.RoundRobin |
round_robin |
Cycle through the providers across consecutive queries |
Invalid strategy values fall back to single. Behavioral details are described in architecture.
WebFetchConfiguration
Constructor
All values are read from environment variables.
Properties
| Property | Type | Default | Env Variable | Description |
|---|---|---|---|---|
maxCharsPerFetch |
number |
10000 |
LLM_CHAT_WEB_MAX_CHARS_PER_FETCH |
Default max chars per URL fetch |
maxCharsPerFetchLimit |
number |
100000 |
LLM_CHAT_WEB_MAX_CHARS_PER_FETCH_LIMIT |
Hard limit for per-request max_chars override |
fetchTimeoutMs |
number |
5000 |
LLM_CHAT_WEB_FETCH_TIMEOUT_MS |
Timeout in ms for URL fetch requests |
maxContentLengthBytes |
number |
10000000 |
LLM_CHAT_WEB_MAX_CONTENT_LENGTH_BYTES |
Max bytes accepted before rejecting the fetch |
maxPdfPages |
number |
100 |
LLM_CHAT_WEB_MAX_PDF_PAGES |
Default max pages extracted from a PDF; a per-call maxPages above it fails the call |
headerRules |
HeaderRule[] |
[] |
— | Host-controlled header rules matched by glob pattern (config only) |
concurrency |
number |
3 |
LLM_CHAT_WEB_FETCH_CONCURRENCY |
Concurrency limit for parallel URL fetches |