We’re building Ever Works in public. Some things may be incomplete, missing, or broken while we continue improving the platform.We’re building Ever Works in public — expect a few rough edges.
Compare · Company builders
Open-source multi-agent runtime for business processes
Aden Hive runs business processes with colonies of agents inside the software you already pay for. Ever Works researches the topic, writes the content, generates the site and keeps it improving on a schedule — content and code in Git repositories you own.
Aden Hive — OpenHive in its open-source form — is an Apache-2.0 Python harness for running business processes with agents: a colony is one persistent Queen, picked from thirteen domain personas, plus as many disposable worker clones as the job needs, coordinating through a shared SQLite ledger with crash-safe resume, budget limits and a human kept in the loop. It is an unusually complete runtime, and the parts of it we do not have are real: fan-out from inside a running turn, large tool results spilled to files behind pointers, an in-turn judge whose critique is fed back into the same conversation, and a process that can be killed mid-task and pick up where it left off. The difference is what exists when the run ends. Hive’s work lands inside systems you already pay for — a CRM row, a sent email, a spreadsheet, a report in the transcript — and it has no publishing target of its own: it can open pull requests, issues and releases against repositories it does not author, but it cannot commit files or trigger a deploy. Ever Works starts from an Idea and finishes with a public artifact: it researches the niche, writes the content, generates the site from a template, deploys it, and keeps re-running that loop on a schedule, with content and code committed to Git repositories you own. Choose on the artifact rather than on the orchestration — one is a very good way to run a process inside your existing SaaS, the other is how you come to own the site that process is meant to feed.
Hive is clear-eyed about where its output goes: effects inside other people’s systems. Its market-research and content personas genuinely research and draft — with web search, news, academic and transcript tools behind them — but there is no publish target at the end, and its Git tools open pull requests, issues and releases against repositories it does not author rather than committing files of their own. Ever Works treats publishable content as the deliverable: the pipeline researches the niche, writes items, categories, tags, collections, comparison pages and SEO metadata, commits them to the Work’s data repository, and deploys the site that renders them. Where we are, precisely — the directory kind is the mature one. Dedicated blog-post and landing-copy writers are still rolling out, so Blog, Website and Landing Page Works are fed by that same items pipeline today.
Both products run unattended, and it would be wrong to suggest otherwise: a colony is schedulable by cron, and its ledger makes a re-run resume cleanly. The difference is what a scheduled run moves. Ever Works’ Scheduled Updates re-run the whole generation pipeline on one of seven cadences from hourly to monthly, retry a failure after fifteen minutes and auto-pause the schedule after three consecutive ones — and the result lands as commits in your repository and a fresh deploy. A Hive re-run advances rows in a colony’s SQLite tracker on one machine, which is exactly right for a process and does nothing for a public page.
Both projects are open source and both self-host, so the licence badge is not the differentiator — and Hive’s Apache-2.0 is the more permissive of the two if what you want is to embed the code. The claim worth making is about feature gating: Ever Works is AGPLv3 with no paid tier of features held back from the open runtime. The sharper difference is ownership of the output. A colony’s own state is a directory of files and a SQLite database on one workstation, unversioned by design — no diff, no history, no review of what the agents produced. Ever Works commits the content, the typed per-Work Knowledge Base and the code to Git repositories you own, gives every Task its own branch and private checkout, and lets a merge policy decide whether an Agent may land its own pull request.
With Hive you start a colony and give a Queen a goal. Ever Works gives you a standing operation: Missions that keep proposing Ideas, Ideas that become Works, and a named, persistent Agent team with scopes, Skills, budgets and inboxes behind it — inside Organizations, Teams and an Org Chart, so the work survives more than one operator. Around that sits governance built for output you publish: quality gates that run build, test, lint and typecheck commands after an Agent finishes and send red work back to it rather than to you, approvals and escalations, and spend caps enforced before the AI call. Ever Works also points both ways at MCP — it consumes external MCP servers and is one, so you can drive Works and Items from your own AI client or the CLI, where Hive is a client only. Store and Company builders are still arriving; today the shipped artifact is the site.
Ever Works vs Aden Hive (OpenHive), head to head.
| Feature | Ever Works | Aden Hive (OpenHive) |
|---|---|---|
| Prompt → working app/site | ✓ | Goal → running colony |
| Researches the topic for you | ✓ | ✓ |
| Writes the content (copy, posts, SEO) | Directory data today; posts rolling out | Drafts only, no publish target |
| Generated code from a template base | ✓ | – |
| Deploys and hosts the artifact | ✓ | – |
| Ongoing autonomous maintenance | Scheduled | Cron-scheduled colonies |
| Content version-controlled in Git | ✓ | – |
| Code in your Git | ✓ | PRs to repos it does not author |
| Named persistent agent workforce | ✓ | Queen + ephemeral workers |
| In-run parallel worker fan-out | – | ✓ |
| Crash-safe long-session park and resume | One-hour session cap | ✓ |
| Open source | ✓ | ✓ |
| Self-hostable runtime | ✓ | ✓ |
| No vendor lock-in | ✓ | ✓ |
Choose Aden Hive when the job is a business process inside software you already run — outbound sequences, CRM hygiene, lead enrichment, inbox triage, warehouse reporting, security scanning — and nothing public has to exist at the end. It is the better buy when the work is fan-out shaped, thousands of independent rows where resume-by-construction over a ledger is the property that matters most; when the team is Python-native and wants a library-shaped runtime it can subclass rather than a platform; and when one operator on one machine, with no accounts and no tenancy, is a feature rather than a limit. It also does several things Ever Works does not: spawning bounded workers from inside a running turn, spilling large tool results to files behind pointers so a long run does not drown in its own context, feeding a judge’s critique back into the same conversation instead of failing the run and starting another, resuming a killed process mid-turn where our agent sessions are capped at an hour, un-parking a run from a reply in Slack or Telegram, planning that is reviewed and approved before execution starts, and prebuilt operations playbooks that are runnable on day one.
Visit Aden Hive (OpenHive) ↗Give Ever Works an idea and its agents research the space, write the content, generate the code, deploy it, and keep it improving — all version-controlled in your own Git.