An agent says it is done; a model says everyone agrees. Both are prose, and prose is not evidence. We build tools where the claim is a thing the system checked: a command that exited zero, an agreement recorded as an event. Two of them are open source and yours to run. 15 more bill per result on Apify.
Real output, not a mock-up. Both transcripts come straight from the repositories, and the first one costs nothing to reproduce.
Ask one model and you get one opinion, delivered with the same confidence whether it is right or not. Ask three and you get three answers plus a job: reading all of them, working out where they actually disagree, and deciding. The disagreement is the useful part, and it is exactly the part you end up doing by hand.
$ npx council --topic "Should the README lead with the product or with the install steps?" \
--participant scripted:…/p1.json --participant scripted:…/p2.json --max-rounds 3
── Turn 1 · round 1 · p1 «Participant 1» · amend → version 1
Arguments: A reader decides in the first screen. Say what the product is and show one real run; install comes next.
── Turn 2 · round 1 · p2 «Participant 2» · amend → version 2
Arguments: Agreed on the order, but the reference still makes the page a wall of text. Move it to docs/ and link it.
── Turn 3 · round 2 · p1 «Participant 1» · agree
Arguments: Moving the reference out keeps the first screen short. I agree.
══ Outcome: consensus
Version 2, author p2
Agreed: p1, p2A coding agent tells you it is done. That claim is prose, and prose is not evidence. Often the work was reviewed by the same model that wrote it, inside the same conversation that produced the mistake — and the plan it is following quietly changed three messages ago.
$ vesna build demo-arithmetic
· building
· T1 building
· T1 review: met, 0 findings
· T1 merged e0c6c92
· T2 building
· T2 review: met, 0 findings
· T2 merged 2790604
· branch review: met, 0 findings
· done
$ echo $?
0Open source, MIT, run on your own machine. Each page shows a real run before it asks you to install anything.
macOS app · CLI · MCP server
Ask several models once. Get one answer they all signed.
Claude, Codex and any OpenAI-compatible model take turns on your problem. Each turn amends one shared solution, accepts it or objects. The session ends when every participant accepts the same version — or tells you plainly that they did not.
Early · macOS on Apple Silicon
CLI · TUI · editor extension
A coding agent with spec-driven development built in.
A request is classified, designed, written up, planned and approved — then built by subagents in parallel git worktrees, each task reviewed by a fresh reviewer that can read and cannot write, each merged only when its review passes. The person approves twice and reads the result.
v0.8 · early, tested, honest about what it does not do yet
Four families of tools on Apify. Every one bills per result, and the colour is just how we group them.
15 tools
Find which hooks and formats beat each competitor’s median, then get original scripts for your brand.
Content mix, cadence and what over-performs for you and your competitors — plus a dated 30-day plan.
Title and thumbnail patterns that win, upload rhythm, topic gaps and new title + thumbnail concepts.
Themes, sentiment over time, spikes with the posts behind them and share of voice — no author profiling.
Google Maps, TripAdvisor, Booking, App Store and Google Play reviews into what to fix first and a reply draft for each.
Which local businesses have no site, a slow one or no online booking — scored, with a short pitch.
Top-ranking pages analysed into an outline, must-cover topics, gaps and FAQ — optional first draft.
Which competitor angles and offers run longest, what is new this fortnight, and fresh ad concepts.
Job posts grouped by company: functions growing, tools in use, buying signals and an ICP fit score.
PDF and image invoices to validated JSON and CSV with line items, tax math checks and duplicate detection.
Every transaction extracted, reconciled page by page against the running balance and categorized.
Transcript with speakers, decisions, action items with timestamps — as Markdown and n8n-ready JSON.
Clean chunks with embeddings, a cited FAQ, topic map and a grounded system prompt for your bot.
Text or image to 4–8 second video, 720p or 1080p, with or without generated sound.
One garment photo into looks on AI-generated models, plus a ready HTML lookbook page.
Most Apify users already collect reviews, posts, ads, job listings or search results. Our tools take that dataset and do the next step.
Google Maps reviews, Instagram or TikTok posts, Facebook Ad Library, LinkedIn jobs, search results — or your own PDFs and recordings.
Point the tool at the dataset in one click. It reads it with your permission only, checks the price up front and starts.
Fix lists, reply drafts, content plans, briefs, JSON and CSV, reports and video files — ready for your sheet, CRM or agent.
Pay per event on Apify: per review, per page, per minute of audio, per second of video. No subscriptions on our side. The two products above are free and open source.
Before any work starts, the tool checks your run budget covers it. If not, it stops without spending a cent.
A file that will not open, a clip that fails our quality check, an empty answer — none of it is charged.
Numbers and quotes are checked against your source. Anything that cannot be found there is flagged or dropped.
Company and content level only. No reviewer names, no personal contact data in the output.
Built and tested, waiting for their publishing slot.
No. Run any tool from Apify Console, its API, a schedule, n8n or Make, or let an AI agent call it through Apify MCP.
On the free plan a tool returns a short preview — an example result and the price table — instead of running. Paid Apify plans run it on your own input.
Results are written to your own Apify run storage and follow your account’s retention settings. Tools only read the dataset you pick.
Yes — tell us which dataset you start from and what the finished result should look like.