adhd is live · free · last verified 8 Nov 2026 · viability monitor
adhd
Open-source parallel divergent ideation for coding agents, fanning out under 15 cognitive frames to beat premature convergence
If your agent keeps returning the same three safe answers on design work, ADHD is a cheap experiment with checkable internals: isolated branches, a hard generator/critic split, and code you can read. The 5/6 win rate over a single-shot baseline is directionally encouraging, but it's an LLM-as-judge eval on six problems at v0.1, so treat the numbers as a signal, not a verdict. On closed-form work, skip it and reach for ToT.
Verdict generatedVerified 5d ago · liveness 60/100 · cite: rightaichoice.com/tools/adhd
- Engineering teams already driving Claude or Codex Agent SDK workflows
- Architecture decisions where the obvious design is the trap
- API and SDK surface design that needs non-obvious alternatives
- Debugging fuzzy intermittent failures that won't reproduce on cue
- Tasks with a single verifiable answer, where extra branches only add latency
- Latency- or token-budget-sensitive interactive agents
- Teams that won't read or adapt the generator/critic prompts
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip ADHD if your task has one verifiable answer and a test you can run, because N divergent branches plus a critic pass only add latency and token spend when there is no range of options to explore.
Every run multiplies model calls — N divergent branches plus a critic pass — so your token bill climbs roughly in proportion to how many frames you fan out
ADHD is published as an open-source method with a preprint, a skill, an npm package and an eval suite on GitHub, so there is no licence fee to budget for. Your real cost is inference: each run fans out N branches plus a critic pass, so compare it against a single-shot prompt on the same model and budget roughly proportional token spend. That makes it cheap to trial on a side project and expensive only where you fan out wide on every request.
In short
adhd — Open-source parallel divergent ideation for coding agents, fanning out under 15 cognitive frames to beat premature convergence. Best for Engineering teams already driving Claude or Codex Agent SDK workflows, Architecture decisions where the obvious design is the trap, API and SDK surface design that needs non-obvious alternatives. Free to use.
What is adhd?
adhd — Open-source parallel divergent ideation for coding agents, fanning out under 15 cognitive frames to beat premature convergence. Best for Engineering teams already driving Claude or Codex Agent SDK workflows, Architecture decisions where the obvious design is the trap, API and SDK surface design that needs non-obvious alternatives. Free to use.
ADHD (Parallel Divergent Ideation) is an open-source inference-time method and reference implementation for LLM coding agents that attacks premature convergence, the failure the paper defines as a model returning the first plausible design candidate and polishing it until the output is competent but forgettable. Udit Akhouri published the v0.1 preprint on 2026-05-25 alongside the code, eval suite, a skill, and an npm package. It's aimed at engineers running agents on open-ended, design-shaped work rather than tasks with a single verifiable answer. The mechanics are a two-phase loop. Phase 1 fans out N parallel branches, each under a structurally different cognitive frame from a library of 15, drawing on regulator, speedrunner, biology, and $0 budget as examples. Each branch gets a generator-only system prompt that forbids evaluation or hedging, and no branch sees another's context, so ideas can't anchor on each other. Phase 2 hands survivors to a separate critic call that scores, clusters, and deepens only the top-K. The generator/critic split is enforced by distinct LLM calls with opposing system prompts rather than promised inside one context, which is the core difference from Chain-of-Thought and the main claim over Tree-of-Thought. On six open-ended engineering problems scored by an independent LLM-as-judge at the same model, ADHD beat a single-shot baseline on five, with mean gains of +5.17 novelty, +4.17 breadth, and +7.67 trap detection on a 0-10 rubric. The intended workloads are architecture decisions, API and SDK design, fuzzy intermittent debugging, refactor planning, and naming. Positioning is deliberate: CoT, ToT, self-consistency, Mixture-of-Agents, and multi-agent debate all optimise for correctness on a closed answer space, while ADHD optimises for the range of non-obvious viable options. It runs on the Claude and Codex Agent SDKs. Pick it when the problem is "what are the options"; pass when there's a test you can run.
Key features
- Parallel divergent ideation across N isolated branches per run
- Library of 15 cognitive frames including regulator, speedrunner, biology, and $0 budget
- No cross-branch context sharing during the divergence phase
- Generator-only system prompt that forbids evaluation, ranking, and hedging
- Generator-critic separation enforced by distinct LLM calls with opposing system prompts
- Critic pass that scores, clusters, and deepens only the top-K survivors
- Vantage-point reframing as the branching driver instead of next-step variation
- Open-source implementation on GitHub with an accompanying eval suite
- Distributed as both a skill and an npm package
- Preprint paper, v0.1, dated 2026-05-25
- Runs on the Claude Agent SDK
- Runs on the Codex Agent SDK
- Evaluated on six open-ended engineering problems
- Independent LLM-as-judge scoring on a 0-10 rubric
- Reported mean gains of +5.17 novelty, +4.17 breadth, +7.67 trap detection
Behind the verdict
We'd reach for ADHD when the deliverable is a set of options and the obvious answer is exactly what you're trying to get away from. Architecture calls, SDK surface design, and that intermittent bug you've failed to reproduce three times all fit: the failure mode there isn't wrong output, it's boring output. The 15-frame library is doing real work here, not cosmetic persona prompting; the paper is explicit that frames are chosen to distort structurally, not to signal expertise. The honest caveat is the evidence base. Six problems, one judge model, v0.1, mean gains on a self-defined 0-10 rubric. That's enough to justify trying it, not enough to justify rewiring a team's process around it. If you need proof, the eval suite ships with the repo so you can rerun it against your own task set. Watch the cost. Parallel branches multiply tokens, and the critic-plus-deepening pass adds more calls on top. On interactive latency budgets this will feel sluggish compared with one CoT trace, and the paper doesn't pretend otherwise. It's the wrong tool for fast autocomplete-style agent loops. The closest alternative is Tree-of-Thought, and the split is clean. ToT branches on next-step variation and searches toward a verifiable answer. ADHD branches on vantage-point reframing and converges on a shortlist. If a unit test can settle the question, ToT wins on cost and precision. If no test exists, ToT's branching driver just gives you nearby ideas. Mechanically, the part we'd verify first is the generator/critic separation, since that's the load-bearing claim and the easiest to accidentally collapse back into a single prompt. The repo and skill make it inspectable. Read the system prompts before you trust the win rate. One gap worth naming: the reported results come from one model
Models under the hood
as of 2026-11-08
Limitations
- ADHD is a v0.1 preprint and open-source method rather than a commercial product, so expect the method and the 15-frame library to keep moving.
- Each ideation run multiplies LLM calls — N divergent branches plus a critic pass — which raises both token cost and wall-clock latency compared with a single-shot prompt.
- The reported evaluation covers six open-ended engineering problems scored by an independent LLM-as-judge, and the paper itself notes that judge may introduce its own biases; treat the novelty, breadth and trap-detection gains as directional rather than settled.
- There is no standalone install; running it requires the Claude and Codex Agent SDKs.
- And it is the wrong shape for closed-form work, where extra branches buy nothing.
as of 2026-10-20
Researching adhd? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is adhd still active?
adhd is live · free · last verified 8 Nov 2026 · viability monitor.
Viability score
How well maintained and how widely used is adhd? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: November 2026
How we score →What's new in adhd
Checked todayAcross the latest 1 update: 1 launch.
What do users say about adhd?
What people actually say about adhd — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
Researched76 mentions across 4 sources (Hacker News, YouTube, GitHub, Lemmy) · researched Aug 29, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +Parallel divergent ideation effectively breaks premature convergence.
- +Mechanical generator-critic separation via opposing system prompts is novel.
- +Library of 15 cognitive frames generates genuinely varied perspectives.
- +No cross-branch context sharing forces true independence in branches.
- +Independent LLM-judge evals show +5 novelty, +4 breadth improvements.
- −One failed LLM call aborts the whole run, hurting reliability.
- −No partial-failure resilience weakens long or complex tasks.
- −Wildcard frame guarantee breaks with --frames 1.
- −Requires Claude and Codex Agent SDK, limiting usability.
- −No native GitHub Copilot support, a common integration ask.
- • Cost of LLM API calls—parallel branching multiplies tokens
- • Setup time for Claude and Codex SDK required
- • Potential cost of troubleshooting reliability issues
How we verified this
Every listed fact about adhd — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 8 Nov 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified adhd 17 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed who it suits, summary, our verdict, features, description, pricing tiers, who should skip it
- — changed setup time, pricing fit, hidden costs, pricing tiers, summary, limitations, who should skip it, switching out, switching in, our verdict, description
- — changed setup time, who it suits, description, our verdict, features, hidden costs, integrations, limitations, switching in, switching out, who should skip it, pricing fit, summary, use cases
- — changed website status
- — changed SSL certificate
- — changed hidden costs, pricing fit, our verdict, setup time, who should skip it, description, limitations, summary, integrations, pricing tiers
Showing the 6 most recent of 17 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does adhd cost in 2026?
adhd: Free to use. Pricing last verified 8 Nov 2026.
Plans compared
Every published adhd tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Open Source | Free | — | Engineering teams already running Claude or Codex Agent SDK workflows who want to test the method on their own open-ended design problems before committing process to itAdds: Starting tier: the full method implementation, the preprint, the 15-frame library, generator-critic separation and parallel branching with convergence, all on GitHub at no licence cost |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where adhd's pricing actually pencils out — and where peers do it cheaper.
ADHD is published as an open-source method with a preprint, a skill, an npm package and an eval suite on GitHub, so there is no licence fee to budget for. Your real cost is inference: each run fans out N branches plus a critic pass, so compare it against a single-shot prompt on the same model and budget roughly proportional token spend. That makes it cheap to trial on a side project and expensive only where you fan out wide on every request.
Has adhd's pricing changed?
No plan price for adhd has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick adhd?
Pick adhd if you are Engineering teams already driving Claude or Codex Agent SDK workflows, Architecture decisions where the obvious design is the trap, API and SDK surface design that needs non-obvious alternatives, Debugging fuzzy intermittent failures that won't reproduce on cue.
Look elsewhere if you are Tasks with a single verifiable answer, where extra branches only add latency, Latency- or token-budget-sensitive interactive agents, Teams that won't read or adapt the generator/critic prompts, Deployments that cannot run the Claude or Codex Agent SDK.
Skip it if Skip ADHD if your task has one verifiable answer and a test you can run, because N divergent branches plus a critic pass only add latency and token spend when there is no range of options to explore.
GeneratedReal-world workflow fit
Concrete scenarios for the personas adhd actually fits — and what changes day-one when you adopt it.
You paste the problem statement into a Claude Agent SDK session running the ADHD skill, fanning out branches under regulator, speedrunner, biology and $0 budget frames with no shared context, then let the critic pass score and cluster the survivors.
Outcome: You get four structurally different designs to argue about in review instead of three variations on the same microservice pattern.
You assign branches different causal frames — timing, resource contention, ordering, external dependency — so each one investigates a different hypothesis rather than restating the first guess.
Outcome: Hypotheses that single-pass prompting tends to miss surface before you start instrumenting, and the critic pass tells you which ones are worth chasing.
You run ADHD over the migration problem and let the critic cluster the top-K survivors into distinct migration paths, then take the clusters into a planning doc.
Outcome: Several non-obvious migration paths with their trade-offs, rather than one consensus plan that everyone quietly doubts.
Use cases
- Generate structurally different architecture options for a new microservice instead of three variations on the same design
- Debug an intermittent race condition by assigning branches different causal frames
- Plan a large refactor with several non-obvious migration paths rather than one consensus plan
- Design an API or SDK surface by contrasting opposing design philosophies in parallel
- Brainstorm naming and positioning that escapes the usual corporate vocabulary
- Generate edge-case test scenarios that single-pass prompting tends to miss
- Produce a range of viable options on interdisciplinary tasks with no single ground truth
Setup time & first value
How long it actually takes to get something useful out of adhd — broken out by persona, not the marketing-page minute.
If you already run the Claude or Codex Agent SDK, expect an afternoon: install the skill or npm package, read the method section of the preprint, and run one problem through a small frame set before fanning out wide. Budget half a day more to tune branch count against your token and latency budget. Coming from plain single-shot prompting, the first hour is mostly unlearning the instinct to ask
Switching to or from adhd
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From single-shot prompting: keep the same model and task, swap the single request for N isolated branches under different cognitive frames plus a critic pass
- →From Chain-of-Thought prompting: keep the reasoning request but move branching from next-step variation to vantage-point reframing and split evaluation into a separate call
- →From Tree-of-Thought: reuse the tree shape but isolate branch contexts and replace the interleaved evaluator with a posterior critic pass
- ↗To Tree-of-Thought: when you need search with backtracking over candidate next-steps on a verifiable answer
- ↗To multi-agent debate: when you want cross-critique rounds to push toward factuality rather than range of options
- ↗To single-shot prompting: when latency and token spend matter more than the range of options you generate
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “adhd”, and we withheld 6: 6 could not be judged, because “adhd” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about adhd.
Official links
What are the best adhd alternatives?
The closest alternatives to adhd we list are Proliferate, Shob and Codex by OpenAI. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Proliferate | Freemium | Pricing not published | — |
| Shob | Free | Free | — |
| Codex by OpenAI | Freemium | Pricing not published | — |
Alternatives to adhd
View all
Proliferate
Parallel coding agents in isolated git worktrees, from one open-source desktop IDE
Shob
Open-source, MIT-licensed terminal workspace for running and orchestrating multiple AI coding agents in parallel.
Codex by OpenAI
OpenAI Codex runs parallel cloud coding agents that write code, run tests, and open pull requests for review
Tools that pair well with adhd
Common stack mates teams adopt alongside adhd, with the specific reason each pairing earns its keep.
Proliferate
Parallel coding agents in isolated git worktrees, from one open-source desktop IDE
Shob
Open-source, MIT-licensed terminal workspace for running and orchestrating multiple AI coding agents in parallel.
Codex by OpenAI
OpenAI Codex runs parallel cloud coding agents that write code, run tests, and open pull requests for review
adhd vs its closest rival
adhd's closest head-to-head is with Praktika — These aren't competitors, so there's no 'pick one' answer — the only question is whether either fits your problem. If you run Claude or Codex Agent SDK workflows and keep getting the first plausible design back, ADHD is free on GitHub and attacks that exact failure mode with framed parallel branching and a separate critic pass. If you read a language fine but freeze out loud, Praktika is the buy: unlimited low-stakes voice reps on your phone with corrections folded into the dialogue. Budget, buyer, and job are disjoint; choose on your problem, not on a matchup. Verdict generated
Head-to-head comparisons
Adhd vs Praktika
These aren't competitors, so there's no 'pick one' answer — the only question is whether either fits your problem. If you run Claude or Codex Agent SDK workflows and keep getting the first plausible design back, ADHD is free on GitHub and attacks that exact failure mode with framed parallel branching and a separate critic pass. If you read a language fine but freeze out loud, Praktika is the buy: unlimited low-stakes voice reps on your phone with corrections folded into the dialogue. Budget, buyer, and job are disjoint; choose on your problem, not on a matchup.
GeneratedAdhd vs Temporal Ai
These are not competitors and you should not be choosing between them. ADHD is a free, open-source ideation method you layer onto a Claude or Codex Agent SDK workflow when the model keeps converging on the first plausible design; it buys you better options, not reliability. Temporal is a paid-or-self-hosted durable execution platform whose entire job is keeping a running agent or workflow alive through crashes, retries, and multi-day waits. If your problem is 'the agent keeps suggesting the obvious thing,' read the ADHD paper. If your problem is 'the agent loses its progress when a worker dies,' that is Temporal, and no branching strategy fixes it.
GeneratedAdhd vs Audioeye
These aren't competitors — they're different categories with different buyers, so pick based on the problem, not the comparison. If your problem is that a coding agent returns the first plausible design and polishes it, ADHD is the free, source-available method, provided you're already running the Claude or Codex Agent SDK. If your problem is that your website fails WCAG 2.2 A/AA and you need continuous monitoring, expert-written fixes, VPAT/ACR paperwork, and legal documentation, AudioEye is the paid managed service. Budget-wise: ADHD costs nothing but latency and tokens; AudioEye costs money but replaces headcount you'd otherwise hire.
GeneratedAdhd vs Appgyver
Choose ADHD if you need a free, open-source method to boost creative coding agent ideation and avoid premature convergence on a single solution. Choose AppGyver if you are already an SAP customer and need a unified low-code/pro-code platform for building extensions, automating workflows, and integrating AI agents into your SAP landscape. These tools serve completely different use cases and hardly compete.
GeneratedFrequently asked questions
Best-of guides
For makers of adhd
- Get a Verified Listing for adhd ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks of the review starting.
Used adhd? Help shape our editorial sentiment research.