Your first creator workflow

Ask for an outcome

Your AI client can coordinate several Mittalmar tools in one conversation. Give it a topic, the result you want, and any constraints that matter.

Find inspiration in your saved research

Keep brainstorming in your AI client and ask Mittalmar for references. Search your favorites and collections together, or name the collections you want to use. Saved videos remain searchable even when they have few views.

If several collections share a name, choose the intended collection before continuing. For only a particular collection, ask for that collection without favorites. Subcollections are included only when requested.

These searches read stored references without using your Mittalmar AI budget or starting generation. Your AI client can use the results to draft alternatives. Search matches saved titles and channel names, or curated Format text; it does not inspect every thumbnail or search video transcripts.

Results include coverage and available metric dates. A shortlist is not a complete collection analysis, and views per subscriber is not a comparison with a channel’s typical performance. Ask for the next page or details of a returned video when you need more evidence. Thumbnail previews depend on your client; image links remain available.

Ask for the creator research memory with your topic to retrieve a compact, relevant brief: your authored direction, dated stored mechanisms, prior research and source citations. The brief reports how much it previews and keeps qualified outcome learning separate from preferences. Ask for detail: full or follow a cited workflow, report or transcript passage when you need deeper evidence. Retrieving stored memory does not start paid research or require CSV uploads.

Investigate datasets in Claude or another AI client

Mittalmar supplies stored evidence and computed analytics; your AI client can explore alternatives, interpret the results and build charts. You do not need to upload CSVs or create a Project to investigate a dataset.

1. Find the available evidence

   Your assistant discovers visible collections, favorites and tracked competitors, then checks their coverage and available observation dates. Collection and competitor lists have separate pagination; a first page is not the whole source inventory. Inventory counts include all video formats, while the optional saved and tracked overviews count eligible medium and long videos. A source-name search filters the inventory, not those whole-scope overviews.

2. Compute comparisons

   Ask for up to four named groups with different topics, channels, publication periods or video lengths. The research subject provides context; explicit group filters determine which videos are counted. Mittalmar computes over every matching stored video in each selected scope, rather than only the videos on a returned page. A video can belong to more than one group; overlapping groups are not independent samples. Groups are read separately and can change between reads; the response reports its read window.

3. Inspect and interpret

   Retrieve the supporting video IDs, original titles, thumbnail links and metric dates. Your assistant can chart returned count series and distributions, investigate exceptions and explain alternatives. A thumbnail link does not mean its image has been analyzed; use acquired source passages when making claims about a competitor’s actual content.

Dataset comparisons and source inspection include medium and long videos with a known duration of at least four minutes, excluding explicit Shorts. Results report their scope, missing information and limitations. Stored topic or editorial classifications are inferred labels, not verified content. Label groups cover only currently classified eligible videos, may overlap and have their own preview limits; they are not counts of the entire market. Target language and target duration qualify pattern comparisons, while explicit minimum and maximum duration filters narrow the descriptive population. Lifetime views from differently aged videos do not establish that one approach performs better; comparable performance conclusions are withheld when the evidence is not qualified.

Discovery, dataset analysis and source inspection read your available stored evidence without changing saved work, starting fresh collection or using your Mittalmar AI budget. Your AI client’s own costs still apply. If evidence is missing or outdated, request new research separately and review its bounded allowance usage.

Use the existing Project linking and research document workflows when you want to save the conclusions. Dataset reads do not automatically choose an angle or alter your Project.

Analyze collections without exporting CSVs

Mittalmar computes across the complete selected stored scope, deduplicating overlapping videos. You can also analyze favorites or the indexed uploads of your explicitly tracked competitors. Your connected channel is excluded from tracked-competitor scope. These reads use no Mittalmar AI budget and do not refresh sources.

Results explain the scope, denominators and available metric observation dates. Channel breakdowns show up to 30 groups with omitted-video counts; older year groups beyond the newest 100 are combined. Lifetime views span different publication ages and observation dates, so their descriptive statistics do not establish comparable performance. Publication frequency does not prove audience demand or a market gap.

Start from a rough video topic

Keep the idea, its research and production choices together in a Project. Supply an existing Project or let a rough topic create one before research starts; follow its Project Research link. Research explores positioning before you choose a direction. After you explicitly choose an angle, title options, thumbnail briefs and Script writing use the current source-bound research automatically when you supply the Project. A changed core premise makes older research stale; stale, unfinished and unselected research does not silently guide generation. Your authored working title and core idea stay intact.

Research focuses on medium and long videos with a known duration of at least four minutes. Shorts and unknown-duration videos are excluded from the evidence used to compare competitors, develop angles and guide production. Older research that does not meet this policy becomes stale; refresh it before choosing an angle.

Start with a rough topic; collections and competitor lists are optional. Mittalmar uses your creator context to discover candidate competitors and adjacent niches, acquire available transcripts and original thumbnail images, compute analytics and propose cited positioning angles. Suggested transfers remain hypotheses; missing source content does not prove a competitor omitted something.

The tool returns the Project Research link and a workflow ID immediately; research continues in the background, attached to the Project from the start. Completed previews prioritize cited source examples, showing up to 12 evidence examples, and report their preview coverage. Ask the assistant to keep monitoring that same workflow until it finishes, compare completed angles and retrieve more evidence or acquired transcript passages. Progress reads can wait up to 25 seconds using waitSeconds: 25; the returned next action tells the assistant to continue monitoring. Research continues independently if the conversation pauses. Source dates, coverage and uncertainty remain visible.

Starting, refining and resuming workflows change saved work without a separate Mittalmar confirmation challenge. Starting or refining research uses your included Mittalmar AI allowance within one bounded job cap. Follow the returned usage receipt; exhausted allowance is reported explicitly. Reading progress, evidence and acquired passages uses no additional AI allowance. Your connected AI client’s own costs still apply.

Project research stays cached until you materially change the core premise or research direction. Repeated requests for the same premise and direction reuse the current version. If you explicitly want newer evidence for that same idea, ask to refresh its completed research. This starts a separate bounded background version using included allowance, preserving the idea, research direction and source scope. Retrying the same refresh request with the same idempotency key reuses its admitted version. New Project research does not schedule daily paid refreshes. For an older workflow with an opted-in refresh schedule, tell the assistant to refine it with autoRefresh: false to stop that schedule.

A decision brief assesses your idea or selected Script against acquired competitor evidence: what overlaps, where it can differ, and specific changes to consider. It identifies the supplied script coverage and cites your actual guidance and original source passages. A separate source review can reject unsupported claims before you use the result; this model-assisted check does not guarantee correctness.

Use mode: existing-sources on the existing research refresh action to review the current draft without requesting fresh discovery. This creates a new bounded analysis using eligible sources already acquired for the same Project idea, retaining their original dates. It reloads your current creator context and selected Script; it does not reuse the old conclusions or edit your manuscript. Fresh discovery remains a separate request and can wait for search availability. Earlier reports that have not passed the current source review remain readable but cannot supply a selected production direction.

Progress reads return a compact decision brief with complete positioning angles, preview counts and deeper read pointers. Request detail: full for complete retained intelligence. Ask for the frozen research context to page through the original creator/Project source excerpts used in that report; Asset IDs, chunk indexes, source coverage and a hash preserve the citation trail. For competitor content, retrieve the acquired timestamped passages before making content claims.

A materially different refinement creates a separate version and can reuse available source evidence. An unchanged premise and direction reuse current research. Resume unfinished research within its original cap, or cancel to stop subsequent work. Usage already incurred remains recorded; completed and cancelled workflows remain available.

Choosing an angle requires confirmation and saves its research in the same Project. You do not need to copy the research or create another Project. If you change the premise or select a different research version while choosing, refresh the Project research before trying again. Older standalone research can still be attached to an owned Project or create one.

The selected research direction supplies context within the Project automatically. If you also request a document export, the confirmed save attaches a Markdown research Asset without generating a Script or using Mittalmar AI budget. It includes findings, selected original references, coverage and limitations. Large snapshots retain workflow references for additional evidence pages; saving does not verify the claims. Identical document content can reuse the existing Asset.

Read or process saved collection analysis

Reading analysis versions and evidence pages is free of Mittalmar AI budget usage. Whole-snapshot counts are computed; topic, relevance and editorial labels are inferred from titles and metadata. They do not verify the actual content or visuals.

To classify a collection, ask the assistant to start a new analysis with a subject and a budget up to $3, then complete the approved run. Starting and retrying create saved versions and require confirmation. The approval returns a continuation receipt bound to that report and its original cap. The confirmed analysis continues automatically in the background. Your assistant can also resume bounded processing with that receipt without another Mittalmar challenge until it expires; otherwise a processing step needs confirmation. Processing may use AI budget within the original cap and your account budget. Read the returned status to monitor analysis and inspect computed findings. Failed records and remaining coverage stay visible; retrying preserves the original version.

A useful first workflow

1. Ground the work in your channel

   Ask Mittalmar to use your channel blueprint so the suggestions fit your audience and creative identity.

   Try askingUse my channel blueprint to summarize who my content is for.

2. Explore and review ideas

   Generate a few distinct angles, then score the ones you might pursue.

   Try askingDraft five video ideas about rebuilding audience trust, then score the strongest two against my channel.

3. Build the package

   Create title options and a matching thumbnail direction before committing to production.

   Try askingPlan a title and thumbnail package for the strongest idea.

4. Confirm any saved work

   One-off saves require confirmation. For ongoing work, link a Project once using the workflow below. Thumbnail generation still requires its own confirmation.

   Try askingSave the approved idea and three titles as a project after I confirm.

Generate narration for a saved script

Narration uses ElevenLabs through your signed-in Mittalmar account. Browse voices and listen to existing samples before confirming a generation. Generating an opening, a manuscript passage, or a full saved script uses your AI budget and saves a take with its text and available word timings. The text must match the saved script; long manuscripts may need several passage takes.

If generation times out, ask the assistant to check the same generation using its idempotency key. Recovery can finish attaching already generated audio and its existing usage record without another provider call or charge. An uncertain provider outcome needs reconciliation before another paid attempt. Retrieve a completed take to get its audio link, valid for 15 minutes; playback depends on your client. Narration does not generate intro music.

Narrate in your own voice

Upload one to five clean recordings of yourself to a Project: one speaker, no music or background noise, ideally one to three minutes in total. After you confirm that the recordings are your own voice, Mittalmar creates your voice through ElevenLabs. Cloning does not use your AI budget; narration with your voice does. You can keep three cloned voices. Delete one to make room for another. Saved takes keep their audio. You can also record and manage your voice in the script editor’s Voiceover view.

Keep drafts in your Project

Choose the intended Project, or give a new Project a name before its idea is settled. Confirm the link once: completed ideas and title batches can then be saved to that Project for 30 days without another Mittalmar confirmation for every revision. Your client may still ask for its own write approval.

Your assistant sends a save after each completed draft and returns a Project link when it succeeds. Mittalmar does not watch the conversation in the background: if the assistant does not call the save tool, the draft remains only in your chat. Saving your client’s text does not use your Mittalmar AI budget.

Revisions update the same draft; distinct title alternatives stay separate. A saved title becomes the Working title only when you choose it. If you edit the same work in Mittalmar, the assistant must reconcile the versions before saving over your changes.

Ask “Stop syncing this Project” to stop further saves without deleting saved work. Link again after the session expires, and choose a new link when switching Projects.

Change your channel direction

Your assistant can preview and save changes to your creative blueprint after confirmation. Each save creates a new version and keeps previous versions available for rollback. The updated direction is available immediately for ideas, titles and thumbnail planning. Creator memory refreshes from saved inputs on its next read. Direction health flags a missing audience or an old or unconfirmed direction.

Create projects from saved stories

A batch can create up to twenty projects with one confirmation. Each project can retain up to fifty structured sources, including social posts and articles. Your assistant can also add sources to an existing project after confirmation. Source excerpts are supplied to new research runs and script preparation, with their original attribution. Saving an excerpt does not verify its claims or fetch additional content from its URL.

Research and write a Script in your AI client

Enable Script saving when you confirm the Project link. Your assistant can save a hook, its supporting context and open loops, an outline, or a complete manuscript. Writing happens in your AI client; preparing instructions and saving the result do not use your Mittalmar AI budget. Your client’s model and search costs still apply.

The writing stages are payoffs, hook recommendations, hook context, hook, outline, and manuscript. Your assistant can retrieve Mittalmar’s instructions for each stage, including your selected voice style, and execute them in the conversation. Read saved Project research first and pass relevant passages, source links and uncertainty into the writing brief. Preparing instructions does not perform web research: ask your client to research before drafting.

Research is saved as a Markdown Asset and becomes readable source material in Mittalmar. Each document save requires confirmation. Different content creates a separate document; identical content is reused. Saving records your evidence without verifying its claims. Read all returned pages when reviewing a source. If a confirmed save reports a failure, retry the same document with a new save request and confirm again.

Editing the same draft replaces its text; it is not automatic version history. To preserve a version, the assistant reads the current Script and creates a named copy with a new draft key. The copy retains the existing content and source selections. Only select the new version when you choose it. If either draft changes in Mittalmar, the assistant must read the latest revision before saving.

Attach files made in your AI client

Upload external thumbnails, video drafts, voiceovers, music, broll or other files to an owned Project without using AI budget. Uploads require confirmation and the file’s MD5 checksum (SHA-256 for file chunks), and return a single-use upload URL with the headers your client must send. Start the transfer within ten minutes, then confirm completion to attach it. Files can be up to 1 GiB, including finished video renders; PNG, JPEG and WebP thumbnails can be up to 8 MiB. Uploads count toward your plan’s Project storage: Starter 10 GB, Pro 50 GB, Studio 200 GB. Deleting an uploaded video, audio or other file frees its space. Uploaded thumbnails keep counting after deletion because generated images can still use them; each is at most 8 MiB. Your client must be able to send an HTTP PUT request to the returned URL. If its network policy blocks the upload host, transfer the file from a client that can reach it.

Attached files have unlisted URLs with random keys: anyone with the link can open it, but it is not listed or indexed, and attaching never publishes anything to YouTube. Files appear in the Project’s Assets tab with download links. Uploaded thumbnails also appear in the thumbnail picker and feed preview. You can replace the selected thumbnail or clear it by passing imageId: null to select_project_thumbnail. Read the Project context to check its assets and selection; thumbnail previews depend on your client’s image support.

Generate and view thumbnails

Thumbnail generation runs in the background. After you confirm the request, Mittalmar returns a batch ID.

When the batch is ready, ask your client to check it:

To keep the generation with a Project, ask your client to link the completed batch after you inspect it. This write step requires confirmation.

Move files with your Project

Attach external thumbnails, video drafts, narration, music and editing references. Confirm one file or a batch, then transfer and complete each upload. If your client cannot fetch upload or download links, ask it to use file chunks through the connection. Upload links support files up to 1 GiB and thumbnails up to 8 MiB. File chunks support files up to 80 MiB; send larger videos through the upload link. Completed files appear in the Project Assets tab; video and audio uploads also work there.

Script updates retain prior content in revision history. You can select an existing title or Script by its ID. Ask your assistant to save publishing descriptions and tags separately from the Project's core idea.

Upload privately, then review

Connect YouTube with publishing permission in Mittalmar. Confirm a private upload with your selected title, thumbnail, description and explicit audience and synthetic-media decisions. You can include narration-timed captions, chapters and a playlist. Uploading continues in the background; ask your assistant to monitor the same Project upload.

The review reports pass, warning or failure with evidence. Checks that need listening, claim-context judgment, footage rights or Studio inspection stay visible. Open the Project's review page, inspect the private video and acknowledge each warning. Publishing or scheduling requires your click there; the connected assistant never publishes unattended. Complete Studio-only options and copyright or ad-suitability Checks in YouTube Studio.

Understand AI usage

Some creative operations use your Mittalmar AI budget. You can check your remaining budget or estimate a packaging workflow before running it.