Call Data API

curl https://ext.trellus.ai/visible \
  -H "api_key: $TRELLUS_API_KEY"

curl https://ext.trellus.ai/bootstrap \
  -H "api_key: $TRELLUS_API_KEY"

curl https://rpt-2.trellus.ai/session-list-v6 \
  -H "Authorization: Bearer $TRELLUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "start": 1790812800000000,
    "end": 1791417600000000,
    "cnf": [
      [{ "column_name": "conversation_type", "operator": "EQ", "value_safe": 1 }]
    ],
    "limit": 50,
    "select": ["session_id", "started_at", "customer_address", "customer_data.name", "customer_data.title", "customer_data.company_name", "turn_data.texts", "turn_data.is_customers", "turn_data.starts"]
  }'

curl https://rpt-2.trellus.ai/metric-details-v7 \
  -H "Authorization: Bearer $TRELLUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "start": 1790812800000000,
    "end": 1791417600000000,
    "selects": [
      { "cnf": [] },
      { "column_name": "duration" },
      { "cnf": [[{ "column_name": "dial_metric.live_duration", "operator": "GE", "value_safe": 30 }]] }
    ],
    "cnf": [[{ "column_name": "conversation_type", "operator": "EQ", "value_safe": 1 }]],
    "group_by": { "column_name": "user_log.disposition" }
  }'

curl https://rpt-2.trellus.ai/session \
  -H "Authorization: Bearer $TRELLUS_API_KEY" \
  -H "x-session-id: SESSION_ID" \
  -H "x-proto-as-json: true"

curl https://api.trellus.ai/get-audio-links \
  -H "api_key: $TRELLUS_API_KEY" \
  -H "session_id: SESSION_ID"

[{ "column_name": "dial_metric.is_inbound", "operator": "EQ", "value_safe": true }]

[{ "column_name": "duration", "operator": "GE", "value_safe": 300 }]

[{ "column_name": "dial_metric.live_duration", "operator": "GE", "value_safe": 30 }]

[{ "column_name": "agent_joined", "operator": "EQ", "value_safe": false }]

[{ "column_name": "resource_id", "operator": "EQ", "value_safe": "USER_ID" }]

[{ "column_name": "resource_id", "operator": "EQ", "value_safe": "USER_ID_FROM_BOOTSTRAP" }]

[{ "column_name": "user_log.disposition", "operator": "IN", "value_safe": ["Interested", "Demo booked"] }]

[{ "column_name": "user_log.purpose", "operator": "EQ", "value_safe": "Follow up" }]

Implement a server-side client for the Trellus Call Data API.
Follow my existing project's language and framework. Implement the integration and give
me setup steps plus a small test. Use these reporting endpoints to read data; do not use
them to manage dialer settings or CRM mappings.

Read TRELLUS_API_KEY from the server environment. Never ask me to paste a key into chat.
If it is missing, implement the code and explain how to set it privately before live requests.
Do not expose it in browser code, URLs, logs, output, or prompts. This prompt has no credential.
Ask about the timezone or team only when my requested report depends on that choice.

Authentication and endpoint map:
- GET https://ext.trellus.ai/visible — api_key header; optional team_id header.
- GET https://ext.trellus.ai/bootstrap — api_key header; key owner's account configuration.
- POST https://rpt-2.trellus.ai/session-list-v6 — Authorization: Bearer <API_KEY>; JSON body.
- POST https://rpt-2.trellus.ai/metric-details-v7 — Authorization: Bearer <API_KEY>; JSON body.
- GET https://rpt-2.trellus.ai/session — Authorization: Bearer <API_KEY>, x-session-id, x-proto-as-json: true.
- GET https://api.trellus.ai/get-audio-links — api_key and session_id headers.

Call /visible first to verify identity and read permissions. A key belongs to a user;
team_id selects a permitted team and does not expand access. Do not assume a separate read-only key.
Use /bootstrap for exact disposition, sentiment and purpose strings.

List requests require start (Unix microseconds) and select (array of column paths).
end is exclusive and defaults to now. cnf defaults to []; limit defaults to 50, maximum 1000.
For calls, cnf = [[{ "column_name": "conversation_type", "operator": "EQ", "value_safe": 1 }]].
Fields include session_id, started_at, resource_id, customer_address, customer_data.name,
customer_data.title, customer_data.company_name, duration, turn_data.texts,
turn_data.is_customers and turn_data.starts. There is no standard customer email column.
Return rows are arrays in select order; started_at is seconds, not microseconds.
cnf is AND of OR groups. Filter values for started_at use seconds. Turn offsets and duration use seconds.
For counts/sums, use metric-details-v7 with selects: [{ "cnf": [] }, { "column_name": "duration" }].
If grouped, the group value precedes numeric cells. Both column_name and cnf can appear in a select.

Request parameter definitions:
- start: inclusive range start, Unix microseconds; required for list and metrics.
- end: exclusive range end, Unix microseconds; optional, defaults to now.
- team_id: optional report-body team selector; defaults to the key owner's team. One team per request.
- select: required list of exact column paths for session-list-v6; response cell i corresponds to select[i].
- limit: session-list row cap, default 50, maximum 1000; not a page size with a continuation token.
- cnf: optional AND of OR filter groups. [] imposes no extra filter; avoid empty inner groups.
- column_name: exact column path to filter, sum or group; not an alias or display label.
- operator: EQ equals; NEQ not equal; GT greater; GE greater or equal; LT less;
  LE less or equal; IN is one of the value_safe array; NIN is not in that array;
  REGEX searches a string using the backend regex engine (no look-around/backreferences).
- value_safe: typed comparison value, not a stringified JSON value. Booleans use true/false,
  numerics use numbers, IN/NIN use arrays, bytes use base64, recognized enum columns accept names or numeric values. user_log.source is a numeric
  field for list results and filters; use its numeric codes, not the explanatory labels.
- reducer: required only when a filter path contains [*]; ANY = at least one item matches,
  ALL = every item matches. Do not attach reducer to a path without [*].
- selects: required metric-details-v7 aggregation array. { "cnf": [] } counts matching rows;
  { "column_name": "duration" } sums seconds; specifying both gives a filtered sum.
  Boolean sums count true values. Each aggregation's cnf is additional to the top-level cnf.
  Empty {} is invalid. Numeric response cells follow selects order.
- group_by: optional metrics object with column_name; its value is the first cell in each row.
  String/boolean/enum grouping uses exact values; float/time grouping requires step_size_safe
  and step_start_safe. Integer grouping can use exact values or steps.
- step_size_safe: bin width; step_start_safe: bin origin; step_modulus_safe: optional repeat period.
  Steps use the grouped column's units (started_at uses seconds). A step requires an origin;
  do not supply step parameters for strings, booleans or enums. Group cells with steps are bin
  indices: floor(((value - origin) modulo period, if supplied) / width), not timestamps.
  Compatibility note: when grouping started_at, the current endpoint scales that bin index by
  1,000,000. Divide the returned group cell by 1,000,000 before reconstructing a time bin.

Response definitions and defaults:
- /visible users: readable user records, including user_id for resource_id matching.
  user_id_to_email maps known user IDs to email addresses; not every user must have an entry.
  team is included only for whole-team readers; absence does not mean the key failed.
  groups are readable group records; memberships link user_group_id to member_id;
  member_is_group distinguishes a group member from a user member.
- /bootstrap: the key owner's account configuration and option vocabulary, not a call list.
- session-list-v6: JSON array of positional row arrays; [] means no matching readable sessions.
  Each cell's type/meaning is defined in the catalogue below. Results are newest started_at first.
- metric-details-v7: JSON row arrays; with grouping, [groupValue, ...aggregationValues].
  Without grouping, cells follow selects order. Counts and sums are numeric, not session objects.
- /session: SessionData object in protobuf JSON, with camelCase property names, enum names
  and base64 bytes. Default scalar/empty fields may be omitted; missing is not permission evidence.
  Catalogue paths are snake_case list-query paths, not property names to copy into detail JSON.
- /get-audio-links: session is one session metadata object; audio is a recording metadata array;
  audio_links[i] is the temporary URL for audio[i]. Empty audio/audio_links arrays may mean
  recording retention expired. Do not treat session as an array.
- Unset proto values can appear as 0, false, empty string or empty arrays in list cells.
  A default value is not proof an event was measured or a prediction/evaluation ran.
  Sales/quality flags are detection signals, not guaranteed ground truth.

Use the exact session_id from a row as x-session-id for /session, not a URL query or JSON body.
/session JSON is camelCase (sessionId, customerData, turnData) and omits wordData.
Inspect 401 text: not found differs from not permissioned and not on db.
Audio returns { session, audio, audio_links }; links expire after one hour.

Pace list requests at least 2.5 seconds apart. The default application limit is 2 requests
per 4 seconds per user, shared with session-list-v5. Metrics: 4 per 4 seconds; detail: 1 per second.
Honor Retry-After on 429, use bounded backoff for transient 5xx, and read error bodies as text.
There is no cursor/offset paging. If a window returns limit rows, split it into smaller date
windows and deduplicate by session_id; never claim a capped response is a full export.
Never log credentials or include them in prompts. Report missing enrichment/transcripts honestly.

Minimal call-list body (compute the dates; do not copy old sample timestamps):
{
  "start": START_UNIX_MICROSECONDS,
  "end": END_UNIX_MICROSECONDS,
  "limit": 50,
  "cnf": [[{ "column_name": "conversation_type", "operator": "EQ", "value_safe": 1 }]],
  "select": ["session_id", "started_at", "resource_id", "customer_address", "customer_data.name",
    "customer_data.title", "customer_data.company_name", "user_log.disposition", "turn_data.texts",
    "turn_data.is_customers", "turn_data.starts"]
}
Use Date.parse(ISO_WITH_TIMEZONE) * 1000 to build start/end in JavaScript.
Use new Date(started_at * 1000) for returned list timestamps. Handle empty results before
indexing a row. Combine the aligned transcript arrays into speaker/text/offset objects.

For errors, preserve the HTTP status and response text but redact credentials and customer
personal data in logs. If correct /session headers still produce not found or not on db,
report that as an unresolved backend detail lookup; do not fabricate a transcript.
Return accurate totals and label capped or sampled results clearly.

Complete field names, types, enum values and definitions:
https://www.trellus.ai/developers/api#fields
OpenAPI starter:
https://www.trellus.ai/developers/openapi.json
Read the field reference before choosing columns. Do not invent aliases.
Member paths containing .[]. are documentation notation. Select the parent array, or use .[0]. or .[*]. projections. Filters through [*] require ANY/ALL; select projections do not use a reducer.