Building a free-tier stack: splitting work across AI tools so none of them runs dry

The question this site is asked most often is not which free tier is best but how far free can go before a card is needed. The honest answer is further than any one tool suggests, if the work is split across tools whose meters do not overlap. A transcriber counts minutes, a voice generator counts characters, an image tool counts tokens that refill every day, and a presentation tool hands out one grant and never refills it. This note lays out how to combine free tiers by the unit each one meters, with figures from the pricing pages behind the 22 reviews on this site, checked October 7, 2026. It is a planning note, not a loophole guide; every limit here is the vendor's own, in the vendor's own numbers.

Sort tools by the clock they run on

Every free tier on this site refills on one of three schedules, and the schedule matters more than the size of the allowance. Daily refills are the most forgiving: Suno gives 50 credits per day, which the help centre says is enough for 10 songs, and Leonardo.Ai gives 150 fast tokens per day; a bad afternoon costs you a day, not a month. Monthly budgets reward planning: Otter.ai provides 300 transcription minutes a month and ElevenLabs 10,000 credits a month, both per their pricing pages as of October 2026, and both are gone until the next cycle if spent in week one. One-time grants are the ones to treat as a test drive rather than a tool: Runway deposits 125 credits once and Gamma 400 credits at signup, and the help centres for both state that free credits do not refresh. Put each tool in your stack under one of those three headings before deciding which one to lean on.

Lean on the tiers that are unmetered where you work

A few free tiers are not rationed on the thing most people use them for, and those should carry the daily load. ChatGPT lists unlimited text chat with the default model on the free plan, per the pricing page checked October 7, 2026, and then marks uploads, images, voice and research as limited without numbers. Perplexity puts its core behaviour, a cited answer with follow-ups, on the free tier. NotebookLM grounds its answers in the sources you upload and caps quantity rather than capability. tl;dv documents unlimited free recording and transcription and rations the AI notes instead. The practical rule: route plain writing, questions and the summarising of pasted text to the tools that do not meter them, and save the metered allowances for the work only a metered tool can do. Most people do this backwards, spending a monthly minute budget on a task that an unmetered chat window would have handled.

Pair tools whose meters do not overlap

The second rule is to avoid stacking two tools that count the same unit, because they will run dry on the same day. Two meeting transcribers both counting minutes is not a stack, it is a double exposure. Otter's 300 minutes come with a 30-minute cap per conversation and 3 lifetime file imports, per the pricing page as of October 2026, so a long meeting hits the per-conversation wall before the monthly one. Fireflies.ai shapes its free plan by storage and AI credits rather than minutes, so the two meter different things and cover each other's gaps. Likewise ElevenLabs counts characters, one credit per character on its V2 multilingual models per the pricing FAQ, while Descript counts media minutes, 60 a month, and its help centre is explicit that a file counts on upload. Narration and editing can sit in different budgets, and a stack built that way fails one piece at a time instead of all at once.

• Transcription: minutes (Otter), storage and credits (Fireflies), unlimited recording with rationed notes (tl;dv).
• Voice: characters counted as credits (ElevenLabs). Editing: media minutes (Descript).
• Images: daily tokens (Leonardo), weekly credits (Ideogram), words rather than numbers (Adobe Firefly).
• Decks and video: one-time grants (Gamma, Runway), best spent on a single finished piece.

Know which allowances you cannot sell

A free tier can be generous on quantity and absolute on rights, and the stack has to respect that. The ElevenLabs pricing page, as of October 2026, lists no commercial licence on the free plan; the Suno review records that its free credits are for non-commercial use. Ideogram's documentation states that free generations are public, and Leonardo lists creations access as public on the free line. Descript exports at 720p with a watermark. None of this is hidden, but it is easy to forget when the allowance is large. If any output in the chain is going to a client, a shop listing or a channel that earns, check the rights line for that specific tool before building the workflow around it. The note on this site about the watermark and commercial-use question goes through those clauses one by one.

Keep the sign-in and data settings in view

Stacking five free accounts means five sets of terms. Ideogram ties its 30 weekly credits to a Google, Apple or Microsoft sign-in; Microsoft Copilot is free with a signed-in Microsoft account; and the Mistral Le Chat review records that the training opt-out on its free plan is a setting the user has to find and switch. A stack built for a small business should use one work email, a password manager, and a ten-minute pass through each tool's data settings on the day the account is created. Dates matter too: the free plan a review describes is the free plan as of the check date, and the Gemini review on this site notes that model access on its free tier was documented to change during October 2026. Re-read the pricing page on the day you depend on it, not the month before.

A worked example at zero dollars

Consider a one-person consultancy that records client calls, writes proposals, and publishes a short narrated clip each month. Calls go to tl;dv, unlimited recording with rationed notes, with Otter's 300 minutes held for interviews under half an hour where the per-conversation cap does not bite. Proposals are drafted in ChatGPT's unmetered text chat and checked against sources in Perplexity, with the client's own documents loaded into NotebookLM so answers are grounded in them. The clip's narration uses part of ElevenLabs' 10,000 monthly characters for an internal draft only, since the free plan carries no commercial licence, and the edit happens inside Descript's 60 media minutes. Gamma's 400 one-time credits build the first deck and are then gone. Nothing in that chain has been paid for, and the first thing that breaks is the narration, which is therefore the first upgrade worth buying.

Bottom line

Free tiers combine well when they meter different things and refill on different clocks. Put the unmetered chat and research tools in the middle, surround them with one tool per metered unit, treat one-time grants as a single project, and read the rights line before any output leaves the building. Every figure above is the vendor's own as of October 2026 and will change; the reviews on this site carry the check date, and the pricing page carries the truth on the day you need it.