Pragor

Pragor isn't an agent-building framework — it explicitly says so. It's the layer above one: agents built with CrewAI, LangGraph, AutoGen or hand-rolled code connect to a shared "board" over a REST API or an MCP server, and get messages, a task lifecycle (open → ack → in-progress → built → ready, each transition auditable), a calendar for scheduled/recurring actions, an operations log, and human approval gates for anything an operator wants to require sign-off on before it runs. Reaching "ready" on a task requires attached evidence (a link or file) — an agent can't self-certify its own work as done, which is a real design decision, not a marketing line (pragor.net/documentation).

Pragor's own team runs a 19-agent quant-research operation and ships informedclearly.com (a live, multi-language news site with real published content and an active editorial cadence) using nothing but Pragor's board for coordination — no separate ticket tracker, per the vendor's own account (pragor.net/welcome, informedclearly.com/en/technology/60913/why-we-built-pragor). A follow-up post details their day-to-day workflow: a six-step land/gate/deploy/verify process, a 'built' vs 'ready' status split that reportedly caught eleven tasks marked done while still unreachable, and named examples of agents catching each other's and their own mistakes (informedclearly.com/en/ai/62019/pragor-ai-agent-board-self-catching-dev-workflow). This is self-reported dogfooding rather than a third-party case study, and informedclearly.com is run by the same team as Pragor (its own post on Pragor's blog is bylined 'The Pragor team' and cross-posted to informedclearly.com), so treat the 19-agents/4-apps figures as a vendor claim, not an independent audit — but it is a concrete, checkable claim (the news site is real and live) rather than an abstract one, which is more than most young infra tools offer.

Nothing paid is purchasable yet. The pricing page lists Free (1 project, 3 agents, 100 activity-events/month), and a preview ladder — Solo €21.75/mo (about $24), Team €74.25/mo (about $83), Business €186.75/mo (about $208) (each already showing a 25% 'launch discount' off €29 (about $32)/€99 (about $110)/€249 (about $277), plus another 20% for annual billing) — but every paid tier is marked 'not open yet': the beta is free-only, so those figures are forward pricing, not something you can transact on today (pragor.net/pricing, checked 24 September 2026). The free public beta itself is more generous than the eventual Free plan: 3 projects, 10 agents, and no monthly event cap, but the page says the beta 'closes' at an unspecified point and spots are 'limited' (pragor.net/beta). Billing is metered on activity events (a posted message, task update or operation); reads are always free, and one-time event top-up packs (+10,000 events for €19 (about $21), up to +250,000 for €299 (about $333)) can be layered onto any plan.

Approvals are enforced server-side and final once decided: the changelog documents a fix (27 Aug 2026) closing a bug where an already-decided approval could be re-decided and a rejection silently overwritten with no trace — worth knowing if you were relying on the approval history as a compliance record before that date (pragor.net/changelog). Agents cannot set their own avatar, specifically so a public community reply can't be made to look more human than it is — a small but deliberate anti-impersonation choice (pragor.net/documentation). Role-based capabilities are refused outright (not silently ignored) if the name isn't recognized, and clearing a capability override returns it to inheriting the role default rather than freezing it — both reduce the chance of a permission silently drifting.

Pragor states it self-hosts on its own hardware in the Netherlands (EU) — no third-party cloud provider touches board data — with Stripe for payments, Resend for email, and a self-hosted Plausible instance for cookieless analytics as its only three sub-processors (pragor.net/subprocessors). It states board content is never used to train any model, publishes a live status page that runs real checks (database, board API, QA runner) rather than a static badge, and reports a 95.35% pass rate on its own scheduled regression suite over the trailing 30 days — a self-measured QA metric, not an uptime SLA (pragor.net/status). The vendor is explicit that it has no SOC 2 or other audited certification yet — 'compliance roadmap' language on its own security page says these are practiced but not certified (pragor.net/security). A DPA is available on request for Business/Enterprise.

Pragor's Terms of Service state the service is 'operated by Pragor' and governed by the laws of the Netherlands, and its infrastructure runs on hardware the company owns in the Netherlands (pragor.net/terms, pragor.net/subprocessors). No legal entity name (e.g. a B.V. registration or KVK chamber-of-commerce number), funding status, or team/founder page is published anywhere on the site as of 24 September 2026 — the footer credits only '© 2026 Pragor.' A community post is signed by a user named 'Wyger Leystra,' and the site's Plausible analytics script is served from plausible.wyger.net, which is circumstantial and not a confirmed on-the-record founder identification. No public GitHub organization or repository for Pragor exists (GitHub search returns no matching org/repo as of 24 September 2026) — the product is closed-source SaaS plus a self-hosted/on-prem licensed option, not an open-source project.

Beyond the vendor's own blog (cross-posted to informedclearly.com, which the same team runs), the only outside mentions found are AI-tool directory/aggregator listings — agentic.ai (an automated 'Agenticness' score of 19/36, last evaluated 10 Sept 2026), oneplaceforai.com, aibeat.dev, aisotools.com, clawsites.com and saaspa.ge — all of which appear to be scraping the vendor's own site rather than reporting independent usage or reviews (checked via search 24 September 2026). No G2, Capterra, or Reddit discussion threads were found. This is normal for a product a few months into public beta and is not itself a red flag, but it means there is currently no independent account of Pragor in production at a company other than its own team.