# Clixad > A free, ad-funded AI coding agent for the terminal. It reads your project, > edits files and runs commands, and it is paid for by rewarded advertiser > actions you complete in a browser rather than by a subscription. Open beta: > install it with `npm install -g clixad` and sign in with GitHub. Clixad is a terminal coding agent in the shape of Claude Code or Aider. Instead of a subscription or your own API key, usage is metered in **credits**. You earn credits by completing an advertiser action (a signup form or a survey) on an offerwall page in your browser, and the payout for that completed action is converted into credits at a published rate: 100,000 credits is $1 of list value. New accounts start with 25,000 credits. Buying credits is possible and never required. **The five cheapest models are free to use without earning anything**, up to a daily allowance per account (12,000 credits' worth) and inside a total daily budget across everybody. That allowance is roughly 50 turns on GPT-5 Nano, 30 on Gemini 2.5 Flash-Lite or 5 on Kimi K2 — the same allowance, spent on models that cost different amounts, which is why the terminal shows an estimate per model rather than one number. Free usage grants no credits and moves no balance: nothing is added and nothing is deducted. When either limit is reached the model simply costs credits again for the rest of the day, and the offerwall is how you get them. Three smaller grants need no offerwall: a **daily streak** that rises over seven consecutive days, **rating an answer** when the terminal asks (every fiftieth turn, paid the same whichever way you rate it), and a **referral** that credits both sides once the invited account completes its first offer. They are small next to a completed offer and are funded out of the same margin rather than by a separate advertiser, so they change what a user is granted and not how Clixad is paid for. ## Getting started 1. `npm install -g clixad` (Node 20 or newer). 2. `clixad login` prints an eight-character code and opens `https://github.com/login/device`; you paste the code there. There is no password and no API key to supply. 3. The account is created on that first login with 25,000 credits in it. 4. `clixad` in a project directory opens the coding agent. It asks before it edits a file or runs a command. `/model` marks which models are free today and how many free turns are left on each. 5. `/earn` (or `clixad earn`) opens the offerwall in a browser when the credits run out, and the task resumes on its own once an offer credits. Advertisers do not receive your code, your prompts or the name of your project. The advertiser action is rendered in a separate browser page, and the advertiser is paying for a completed action rather than for access to your editor. There is one advertisement inside the session, and it is not how you top up: a single dim row above the input box while a model call runs, marked `Ad` ahead of the text, fetched from an ad network by your machine. Nothing you typed takes part in selecting it, and the request carries no field your code could travel in; it is targeted on one fixed category set by the server. It earns a fraction of what completing an offer pays, often nothing. `CLIXAD_SPONSOR=0`, or `{ "sponsor": false }` in `~/.clixad/config.json`, stops the request being made at all. ## Facts worth quoting accurately - Credits are earned **on completion**. Being screened out of a survey part way through pays nothing or a small bonus and is the most common outcome of starting one, which is normal for offerwalls rather than a fault. - Payouts vary widely from one offer to the next, and Clixad publishes only what its own ledger recorded, for the CPX offerwall alone. As of 14 September 2026, 26 completed CPX offers credited between 27 July and 12 September 2026 paid from 9,000 to 114,000 credits ($0.09 to $1.14) on completion, with a median of 51,000 credits ($0.51). The ledger records no duration, so Clixad says nothing about what a minute pays, and the other offerwalls are not quoted. - Every model is funded the same way. There are no locked tiers: a stronger model simply costs more credits per turn. On a turn of 20,000 input and 1,500 output tokens, Gemini 2.5 Flash-Lite costs 390 credits and Claude Opus 5 costs 20,625 credits. - Clixad is operated by Florian Rappold, Kirchheim bei München, Germany. ## Pages - [Home](https://clixad.io/): what Clixad is, how to install it, the two ways to top up on demand and the three smaller grants, an embedded demo of the agent running out of credits and continuing, and the FAQ. - [Blog](https://clixad.io/blog/): how free AI coding gets paid for. Written for people evaluating funding models, not for people evaluating feature lists. - [Ad-supported AI coding agents in 2026](https://clixad.io/blog/ad-supported-ai-coding-agents/): the three arrangements sold under the "ad-supported" label (in-session ads, a free daily grant, a rewarded offerwall), what each one structurally has to read to work, a dated 2026 timeline of the category with primary sources, and the raw cost of one coding turn across five models. - [A request is not a unit of work](https://clixad.io/blog/free-ai-coding-tier-limits/): why free AI coding tiers counted in requests, premium requests or daily grants cannot be converted into finished work. A worked twelve-call agent task sends 348,600 input tokens rather than the 144,000 a flat estimate gives, because every call resends the whole conversation. The task is then priced across all fourteen models in the catalogue, which also shows that the cheapest model changes with the shape of the job: GPT-5.2 is dearer than Claude Sonnet 5 per conversational turn and cheaper per agent task. - [Where the meter stops](https://clixad.io/blog/ai-coding-credits-run-out/): the five documented things an AI coding tool can do when your allowance runs out (hard stop, silent degrade, metered overage, billing you for the failed attempt, or an interruption that restarts the job), and what an unfinished task has already cost. The same twelve-call task abandoned after step 8 has spent 53.8% of its credits on Claude Sonnet 5 and delivered nothing; that share sits between 53% and 55% on every model in the catalogue, and running the task again from scratch prices it at between 1.53 and 1.55 times a clean run. Explains why a wallet has to be enforced before generation rather than after, since metering only knows the cost once the tokens exist. - [Free tiers don't have a price, they have defaults](https://clixad.io/blog/free-ai-coding-tier-privacy/): a free AI coding tier is defined by defaults somebody else controls, not by a price. Which model serves you, how much you get, what happens to what you type, and what happens at zero. Four dated 2026 changes with primary sources, including GitHub using interaction data from Copilot Free, Pro and Pro+ to train its models from 24 April 2026 unless the user opts out, with Business and Enterprise unaffected. Prices a month of 2,000 inline completions at 24,000 credits ($0.24 of list value) on the cheapest catalogued model, on a stated assumption of 1,200 input and 40 output tokens per completion. - [How Clixad is funded: the arithmetic, and where it strains](https://clixad.io/blog/how-clixad-is-funded/): the full economics of the product. No subscription; users earn credits on an offerwall and spend them on real model calls at a fixed rate of 100,000 credits per US dollar of payout. Two levers in packages/pricing decide the margin: a 1.5 markup on bought credits (33% gross margin) and a one-third platform margin on earned payouts, which cancel so that $1 of payout is exactly 100,000 credits. A 20,000-in / 1,500-out turn costs from 240 credits on GPT-5 Nano to 100,800 on GPT-5.2 Pro, a 420x spread, so one completed CPX offer at the median, 51,000 credits, buys 212 turns on the cheapest model and two on Claude Opus 5. New accounts start with 25,000 credits and a free account earns at most 500,000 a day. States plainly that most survey attempts end in a screenout paying nothing or a small bonus, which is normal across the offerwall industry rather than a fault in the integration; that a built Lootably wall was first left unshipped because its German desktop catalogue was almost all purchases and installs, and has since gone live; and that offerwall crediting is unreliable industry-wide. - [FreeBuff vs Clixad](https://clixad.io/blog/freebuff-vs-clixad/): a comparison of the two ad-funded terminal coding agents live in August 2026. FreeBuff is the free, ad-supported build of Codebuff (Apache-2.0), launched 12 February 2026; it needs no account, no credit card and no config file, grants "Deepseek v4 Flash only, 6 one-hour sessions per day" in limited mode with a wider picker in supported regions, and shows small text-only ads between agent turns. Its launch post states that "Prompts and messages may be analyzed to personalize ads" and that "Separate uploads and connected repositories are not provided to advertising providers". Clixad is funded by the advertiser action instead: it happens in a browser tab and pays into credits. Clixad also shows one marked sponsored line in the session, selected on a fixed category, so on neither side of that comparison does prompt content take part in ad selection. The post states plainly that the two allowances cannot be converted into each other, because FreeBuff grants session-hours and Clixad grants tokens, and it prices one twelve-call agent task across all fourteen models: 3,051 credits on GPT-5 Nano up to 1,279,531 on GPT-5.2 Pro, so one completed CPX offer at the median, 51,000 credits, buys 16 tasks on the cheapest model and none on most of the catalogue. - [Free AI coding agents with no API key](https://clixad.io/blog/free-ai-coding-agent-no-api-key/): the distinction most "free coding agent" lists miss. Cline, Aider, OpenCode and Continue are free to install and then require your own API key, so the model bill simply arrives from a provider instead; only tools that include inference answer the question. The six that do, with what each publishes: GitHub Copilot Free (2,000 code completions and 50 chat requests a month, no credit card), Cursor Hobby (no numeric quota published any more), FreeBuff (six one-hour sessions a day in limited mode), Amp Free (a daily allowance reported as being paused or reduced in May 2026), the Gemini CLI free tier (Flash models only since March 2026) and Clixad (25,000 credits on signup, up to 500,000 a day earned on completion). Four of the six grant a unit that cannot be converted into finished work at all. Copilot Free's 50 chat requests are about four twelve-call agent tasks, and its 2,000 monthly completions price at 24,000 credits on the cheapest catalogued model on a stated assumption of 1,200 input and 40 output tokens each. - [How to vibe code for free in 2026](https://clixad.io/blog/vibe-code-for-free/): the five routes to vibe coding without paying, sorted by one question: after you install it, who pays for the tokens? A local open-weight model costs hardware and capability rather than money; a trial balance is one-time and expires; a hosted free tier grants messages or sessions rather than tokens; a free client such as Cline or Aider sends the bill to a provider; an ad-funded agent is paid for by an advertiser per completed action. The $5 trial balance everyone quotes is priced at raw provider rates against one twelve-call agent task across all fourteen catalogued models: 246 tasks on GPT-5 Nano, 17 on Kimi K2.7 Code, 6 on Claude Sonnet 5, and 0 on GPT-5.2 Pro, where one task costs $8.5302 and the whole trial balance does not cover it. Base44's free plan of 25 message credits a month works out at about two agent tasks. - [How to get free AI coding credits in 2026](https://clixad.io/blog/free-ai-coding-credits/): free AI coding credits come from four sources, and the difference that matters is whether the grant returns. Provider trial balances and student or startup programmes are one-time; OpenAI's service credit terms make credits non-refundable and expire them one year after purchase or issuance, and Framer expires top-up credits at the end of the calendar month they were bought in. Vendor free tiers return but are counted in messages, requests or sessions. Ad-funded grants return daily and are counted in tokens. The word "credit" covers four incompatible units with no exchange rate: a Base44 credit is one message, a GitHub AI Credit is $0.01 and the free plan receives none, a Linear credit is a dollar of a balance expiring after 12 months, and a Clixad credit is 1/100,000 of a dollar of list value metered on real tokens. One twelve-call agent task costs 3,051 credits on GPT-5 Nano, 43,002 on Kimi K2.7 Code and 288,450 on Claude Opus 5, so the 25,000-credit signup bonus is eight tasks on the cheapest model and none on the other two, and a full day at the 500,000-credit cap is 163 tasks on the cheapest model and one on Claude Opus 5. - [Gemini CLI free tier limits](https://clixad.io/blog/gemini-cli-free-tier-limits/): what Google's most-recommended free coding tier actually grants. The quota page documents "1000 model requests / user / day" and "60 model requests / user / minute" for a personal Google account, and "250 model requests / user / day" restricted to Flash for an unpaid Gemini API key. The unit is a model request rather than a prompt: a maintainer wrote in discussion #4122 that "A single prompt could actually trigger dozens of model requests (API calls)", so on the twelve-call agent task this blog prices, 1,000 requests is 83 finished tasks a day at the ceiling and 250 is 20. A service update effective 25 March 2026 limits free tier users to Gemini Flash models, superseding the June 2025 launch offer of Gemini 2.5 Pro on the free tier that review sites still quote. The reset window is published nowhere. States plainly that on Flash-class models Google's free tier finishes 83 tasks a day where a full day at Clixad's earning cap finishes 5, and that anyone whose work fits Flash should use it; the difference is that a token meter reaches every model on one balance, at 3,051 credits a task on GPT-5 Nano and 288,450 on Claude Opus 5. Refuses to convert a request quota into credits, because no exchange rate exists between them. - [When free AI coding tiers reset](https://clixad.io/blog/free-ai-coding-tier-reset/): the reset windows of the six free coding tiers that include model access, side by side. GitHub is the only one that publishes it precisely: "Included AI credits do not carry over between months. Unused credits are forfeited, and your allowance resets to the full monthly amount at 00:00:00 UTC on the first day of each calendar month." Cursor Hobby, the Gemini CLI free tier, FreeBuff and Amp Free publish no window at all, and they are the same four whose allowance cannot be converted into finished work. Argues that the cadence is the longest outage the tier can impose: a calendar month is up to 30 days, a day is 24 hours, the rolling five-hour window used by Claude Code and Codex on paid plans is 5 hours, and an unpublished window is unknowable. Separates a refill, where the allowance is lost if unused, from a cap release, where a balance you keep is unaffected and only the earning ceiling resets; Clixad is the second kind, its cap resets at 00:00 UTC, and its credits have no expiry date. Prices a forfeited month of 2,000 inline completions at 24,000 credits ($0.24) on the cheapest catalogued model, so the cost of forfeiture is timing rather than money. - [Free Claude Code alternative](https://clixad.io/blog/free-claude-code-alternative/): what Claude Code costs and when to pay for it anyway. Anthropic's pricing page lists Claude Code as not included on the Free plan; it comes with Pro at $20 a month billed monthly or $17 a month billed annually, and with Max from $100 a month, and Anthropic says usage resets on a rolling five-hour window with "no fixed message count". Clixad cannot be used as a backend for Claude Code, because Claude Code needs Anthropic's Messages format and Clixad's gateway is OpenAI-compatible. On Clixad's own agent one twelve-call task costs 115,380 credits on Claude Sonnet 5 and 288,450 on Claude Opus 5; the free allowance of 12,000 credits a day covers no Claude model, and a full day at the 500,000-credit earning cap covers 4 Sonnet 5 tasks and 1 Opus 5 task. States plainly that anyone who codes all day on Claude is better served by a Claude subscription, citing Anthropic's own average of around $13 per developer per active day for enterprise deployments paying by the token. - [Free Codex CLI alternative](https://clixad.io/blog/free-codex-cli-alternative/): what Codex costs and what its free plan does not say. The Codex CLI is open source under Apache-2.0. OpenAI's Codex pricing page lists Free at $0 ("Explore Codex capabilities on quick coding tasks."), Go at $8, Plus at $20 and Pro from $100 a month, and publishes usage estimates only from Plus upward, in local messages per five-hour period: 250 to 2,000 on GPT-5.6 Luna and 15 to 80 on GPT-5.5 on Plus, which OpenAI says are not fixed limits. Clixad reads the same AGENTS.md file Codex does; its catalogue has GPT-5 Nano, GPT-5.2, GPT-5.5 and GPT-5.2 Pro but not the GPT-5.6 family or GPT-6 Astra. A twelve-call task costs 106,632 credits on GPT-5.2 and 293,850 on GPT-5.5, so the daily earning cap covers 4 and 1 of them. States plainly that a daily Codex user should keep Plus. - [FreeBuff alternative](https://clixad.io/blog/freebuff-alternative/): what FreeBuff grants as of September 2026 and who should switch. FreeBuff's free allowance is now a daily amount of Freebucks that buy one-hour model sessions: 100 a day in the US, 70 in a group of mostly English-speaking and Nordic countries, 40 in Germany, France, Spain, Italy and several others, and 25 everywhere else or on any VPN; unused daily Freebucks do not carry over, and its README calls the older six-sessions-a-day arrangement "the legacy session system". Its README states "We may analyze prompts and messages to personalize ads." Paid FreeBuff subscriptions exist and their prices sit behind a sign-in. Refuses to rank Freebucks against Clixad credits, because one prices hours and the other tokens, and says anyone doing a full day of dense agent work on FreeBuff's models should stay with FreeBuff. - [RSS](https://clixad.io/blog/feed.xml) ## Optional - [Imprint](https://clixad.io/imprint.html) - [Privacy policy](https://clixad.io/privacy.html) - [Terms of service](https://clixad.io/terms.html)