arcmira

Generated from the API document of 2026-10-08, after the on-demand pause. No type or method changed.

• The API's 402 on_demand_paused arrives as PaymentRequiredError with error.type quota_exceeded and no error.unlock. The account's included credits are used and on-demand usage is paused after a declined payment. An admin of the account updates the card at https://arcmira.com/dashboard/billing, and on-demand usage resumes once the invoice is paid.

Generated from the API document of 2026-10-08, after the team account changes. This release removes types and a method, so it is a minor bump while the package is 0.x.

Breaking changes from 0.4.

• trackers.create is removed, with its request types. POST /v1/trackers answers 410 tracker_create_retired. Follow ids or exact names in a monitor with monitors.entities.add(monitor_id, entity_ids=[...], names=[AddEntitiesRequestNamesItem(name=..., type=...)]), which reports the plan limit per entity as reason="tracker_limit_reached".
• Tracker no longer has notify_email, notify_webhook, notify_slack, webhook_url, slack_channel_id or slack_integration_id, and trackers.update no longer takes them. A tracker alerts through its monitor's delivery; set it with monitors.update. Tracker.monitor_id is always set.
• monitors.create no longer takes team_id. Monitor no longer has team or muted, and MonitorTeam is removed. A recipient status is never muted.
• MeResponse.usage.credits.purchased is removed.

Added.

• MeResponse.account (id, name, kind personal or team, plan) and MeResponse.role (owner, admin or member) say whose credits the key spends. email_masked is the person calling.
• The API's 402 member_limit and 402 fair_use_cap arrive as PaymentRequiredError with error.type quota_exceeded and no error.unlock.

Generated from the API document of 2026-10-04, evening, whose descriptions lead with credits. MeResponse.period_resets_at is a string on every plan.

Generated from the v1 document of 2026-10-04, which prices every read in credits. PremiumQuote replaces the TranscriptPurchaseQuote type.

Generated from the v1 document of 2026-10-02. The document dropped from 88 operations to 36. Routes that left it still serve over HTTP, but the SDK no longer has methods for them.

Added.

• client.monitors.entities.add(id, entity_ids=None, names=None, person_match_mode=None) follows entities in a monitor by id, or by exact name and type for a name not yet indexed. It reuses the account's tracker for each entity or creates one, then attaches it. Each id or name gets one result, and one that cannot be followed comes back with attached: false and a reason.
• trackers.create and names accept org for organization, and the duplicate check ignores case.
• client.integrations.slack.list() lists the account's active Slack workspaces with the slack_integration_id and slack_channel_id values a monitor needs for Slack delivery. Slack is connected in the dashboard, not through the API.
• monitors.create takes team_id. Monitor carries access, muted and team.
• feedback.submit takes category and mcp_call_id, and type="experience" reports how a task went as a whole.
• List bodies for mentions, recommendations, search, counts and channel videos echo the applied date window as window: { after, before }.

Breaking changes from 0.3.

• Premium is one read. transcripts.get(video_id, quality="premium") answers ready (200) when the account owns the transcript. Otherwise it starts transcribing the whole video, using credits from the plan and then the account's on-demand budget, and answers pending (202) with the job and a Retry-After header. Read again after Retry-After. Repeated reads join the same job and never use credits twice. When the last transcription of the video failed or was refunded, the read answers failed (200) with the job and last_attempt and uses no credits; pass retry=True to transcribe it again, which uses credits again. TranscriptResult is now TranscriptResult_Ready | TranscriptResult_Pending | TranscriptResult_Failed. Priced refusals carry one RefusedQuote type. The preparation_required state and TranscriptResult_PreparationRequired are gone.
• transcripts.prepare_and_wait is removed, along with PreparationError, PreparationFailedError, PreparationTimeoutError and PremiumUnavailableError. Loop on transcripts.get(..., quality="premium") until state == "ready". The README has the loop.
• transcripts.request and transcripts.status are removed, with the TranscriptRequestSubmitResponse type. The Premium read starts and reports its own job. transcripts.list_requests still lists past Premium transcriptions.
• A Premium refusal raises from the read itself. PaymentRequiredError (402) carries quota_exceeded or spend_limit_exceeded, and ForbiddenError (403) carries paid_plan_required. Nothing is charged.
• Error extras moved under error.details. body.quote is now body.error.details.quote, body.existing_request_id is body.error.details.existing_request_id, and body.existing_id (409 tracker_already_exists) is body.error.details.existing_id.
• Reads take ids. mentions.list and recommendations.list require entity_id (ent_N), and channel_id takes a YouTube channel id (UC plus 22 characters). entity_name, entity_type and channel_name are gone. A name where an id belongs raises BadRequestError with code id_required and names the parameter. Resolve names first with entities.resolve.
• Dates are after and before, half-open [after, before), on every dated read. mentions.list and recommendations.list replace date_from and date_to. date_to was an inclusive day, so date_to="2026-09-01" becomes before="2026-09-02". transcripts.search, mentions.count and channels.videos.list replace published_after and published_before.
• recommendations.list takes class_ (sponsored, organic or mention; omit it for all) in place of mention_class (ad_read, endorsement, mention or all). Recommendation and RecommendationEnrichmentItem carry class_ in place of mention_class, and so does WrongClassificationChange in feedback corrections.
• transcripts.search calls GET /v1/search. Its kind takes sponsored, organic or mention in place of mention, recommendation_sponsored and recommendation_organic. The response renames requested_k to limit and returned_n to returned, and filters.published_after and filters.published_before move to window. TranscriptSearchChunk.text_withheld is gone.
• List bodies name their collection. MentionListResponse.data is mentions, RecommendationListResponse.data is recommendations, and AlertListResponse.data is alerts. Iterating a pager is unchanged.
• Integer ids and MM:SS strings are gone. Mention drops appearance_id, start_timestamp and end_timestamp, and MentionMedia drops id. Recommendation drops recommendation_id, start_timestamp and end_timestamp. Use start_seconds, end_seconds and media.video_id. Alert replaces media_id and appearance_id with video_id. Entity.numeric_id is gone. Premium speakers[].entity_id is an ent_N string, where 0.3 had an integer. Feedback ids are fbk_N.
• monitors.update takes paused in place of is_paused, and is_collapsed and sort_order are gone. Monitor, Tracker and monitor tracker rows carry paused in place of is_paused, and Monitor drops is_collapsed and sort_order.
• Models use the snake_case wire names. Attribute names were already snake_case in 0.3, but the camelCase aliases (videoId, watchUrl, invitationStatus and others) are gone, so model_dump(by_alias=True) and the JSON on the wire now match the attribute names.
• MentionCountsResponse replaces published_after and published_before with window. FeedbackCorrectionResult drops new_mention_class, previous_mention_class, recommendation and rows_affected.
• feedback.submit no longer requires query. type stays required.
• trackers.create follows a channel by its YouTube channel id in entity_name. A channel name raises id_required.
• me.usage.hits is gone. Monitor alerts cost credits now and count in usage.credits.

Removed methods and their replacements.

0.3	0.4
entities.search, entities.lookup, entities.cards	entities.resolve, then entities.get
entities.mentions.list(id)	mentions.list(entity_id=id)
entities.recommendations.list(id)	recommendations.list(entity_id=id)
people.get, topics.get, organizations.get, products.get, channels.get	entities.resolve, then entities.get. For a channel, channels.coverage and channels.videos.list
people.appearances.list	mentions.list(entity_id=..., is_appearance=True)
channels.related.*	mentions.count(channel_ids=..., entity_types=...)
channels.guests.list	mentions.count(channel_ids=..., entity_types="person", mode="appearances")
people.related.*, topics.related.*, organizations.related.*, products.related.*	No direct replacement. mentions.list(entity_id=...) gives the episodes, and mentions.count(video_ids=...) counts what else they mention
transcripts.captions	transcripts.get(video_id). languages lists every caption track and language selects one
transcripts.request, transcripts.status, transcripts.prepare_and_wait	transcripts.get(video_id, quality="premium"), read again on 202
corrections.*, transcripts.edits.*, transcripts.speakers.*, transcripts.merges.*	No SDK method. Report a wrong row with feedback.submit
team.members, team.spend, team.usage_events.list	None. The routes are deleted

The first usable SDK, replacing the URL-only placeholder. The public URL constants remain available.

• Generated synchronous Arcmira and asynchronous AsyncArcmira clients with typed responses and cursor pagination.
• transcripts.get returns a union discriminated on state. The Premium purchase is one job, created by transcripts.request and polled with transcripts.status.
• transcripts.prepare_and_wait reads Premium, posts the purchase once when needed, polls at the pace the API sets and returns the transcript.
• ApiError text leads with the status, code and message.
• Licensed under Apache-2.0.