FieldDIN your field. part of the DIN family.
Get DIN ☰Menu
Get DIN

Three doors, one product

Three doors, one product. Run the free build today, buy more of the thing that is already working for you, or wait for the source, which is not published yet.

Open source

Apache-2.0 when the source is published; binaries today

For the person who wants to see how it works before they trust it with a record.

The plumbing is what will be open: the record format, the honesty vocabulary (the grades, the verdict precedence, the citation contract), the capture contract, the din CLI and the importer's front door. That is the part you would need to read to believe any of the rest, so it is the part that will be published. Today you can run it, not read it: the binary is free, and the Community Edition source is coming.

  • Bring your own prompt. The interpretive contract is a file seam: point din-import --prompt <file> at your own and yours is used instead of ours.
  • Apache-2.0 when the source is published; binaries today. The source is not published yet. Once it is out, you may build on it, ship it and sell what you build.
  • The grading engine, the multi-source fan-in and the roll-ups are not in this repo. They are a compiled binary, not a flag you would find and flip.
The Community Edition source repo is not published yet, so there is no link here rather than one that goes nowhere. It will be Apache-2.0.
Free

Just let me use it

The working tool, compiled. No signup, no card, nothing to configure first.

Install it and you have your own field and this console over it. Point it at one source and it imports your history: a claude.ai export is placed ungraded at no cost; every other source is graded as it is imported, up to the Reader level in the fixed 40,000-character window. Grading runs on your own inference: your assistant CLI (Claude Code or Codex), your own API key (Anthropic, OpenAI, or any OpenAI-compatible endpoint) or a local model (Ollama, vLLM, llama.cpp). On Codex or an OpenAI key, gpt-6-sol at medium effort reads at Reader on Free; din grade on --effort sets the effort, and Analyst and up are paid. The Codex, key and local paths are built and tested on stand-ins; none of them has been validated against a real vendor yet. Live capture of new work, from Claude Code, Cursor, OpenCode or Codex, is a paid feature. One field, up to 3 people and 1 AI agent. What leaves your machine: din import sends your history to the model you chose to grade it (on your own sign-in or key; a local model sends nothing off the machine). DIN never pays for or sees those calls. A paid install also contacts our licensing service, with a random install id, when you run din activate or din activate --renew and at most once a day when the field or the console starts (or with din autostart); a signed check-in lasts 14 days. A free install never checks in. The paid dependency watch, only if you switch it on, contacts DIN's library catalog and the package registries. Nothing else calls out.

One line, Linux x86-64
curl -fsSL https://dinctrl.com/install | sh
Pick the agent
curl -fsSL https://dinctrl.com/install | sh -s -- --agent cursor
Or by hand, reading the archive before you run it:
Linux x86-64, by hand
curl -fsSLO https://dinctrl.com/download/din-1.0.42-linux-x86_64.tar.gz tar xzf din-1.0.42-linux-x86_64.tar.gz && cd din-1.0.42 sh install.sh
Verify it first
curl -fsSLO https://dinctrl.com/download/SHA256SUMS -O https://dinctrl.com/download/SHA256SUMS.minisig minisign -Vm SHA256SUMS -P RWRYpMRMG6vJNhzPUkYbbwTX9ebqgd1rW50Rlkv+/BZ4uAPzfda0KTqK sha256sum --ignore-missing -c SHA256SUMS
Linux and macOS download here. macOS
Download for Linux x86-64 DIN 1.0.42 · Linux x86-64 · .tar.gz · 34.9 MB
The one line downloads the same archive, checks its sha256 against SHA256SUMS (and the signature on it: with minisign when it is installed, otherwise with OpenSSL 3) and only then runs the installer inside it. Or download the archive, read what is in it, and run it yourself. The email is optional and nothing is gated behind it: the download is a download.
Paid and hosted

I want more than one field

Several agents on one field, more fields, and the paid panels. Or, later, we host it.

The free build is one field and history import. Paid is the same download plus a key: live capture, every reading level and the whole-chat read, several agents fanned into one field, more fields, and the paid panels (graded recall, the trust-weighted blast radius, the org roll-up, and team review). Hosted is set up with you: an isolated tenant on our hosted service, with a store cap per org: 10 GB on Starter, 25 GB on Freelancer, 50 GB on Team, 100 GB on Small Business.

  • The meter is fields. Not seats, not volume, and never inference: your model calls run on inference you already have (your assistant CLI, your own API key or a local model) on every tier.
  • Self-host is the default and hosting is never forced.
  • On a self-hosted install, your records are never deleted on a downgrade or a lapse: they are files on your machine.
From $29 a month self-hosted. No card is typed on this page: the checkout is a page on Stripe's own domain, and this console holds no price id and no key.
Every amount on this screen is formatted from the one ladder below, so no figure here can drift from the published one.

Free, paid and hosted, side by side

What each way of running DIN includes
FeatureFreePaid, self-hostedHosted
Can I get it today?Yes: a download, no signup, no cardStarter to Small Business are on sale; for the rest, contact usContact us: set up with you, no card
The program (binary)FreeThe same program, plus a licence keyWe run it for you
The source codeNot published yetNot published yetNot published yet
Where your records liveYour machineYour machineAn isolated tenant on DIN's hosted service
Fields16 to 90 by tier, more quotedBy tier, as self-hosted
People and AI agents on one fieldUp to 3 people and 1 AI agentUp to 25 nodes (people, agents, machines)As self-hosted
Grading, placement and recallIncludedIncludedIncluded
Graded history importIncludedIncludedIncluded
Live capture from your agent (Claude Code, Cursor, OpenCode, Codex), and the grading of those live sessionsNot includedIncludedIncluded
Reading depthUp to Reader: Scanner and Reader, in the fixed 40,000-character windowAll five levels, Scanner to Scholar, depending on your model, and the whole-chat readAs self-hosted
Trust marks and model provenance on every recordIncludedIncludedIncluded
The trust-weighted blast radiusNot includedIncludedIncluded
The dependency watch (scheduled library checks, upgrade briefs, the suspicious-package guard, the org dependency inventory)Not includedIncludedIncluded
Graded recallNot includedIncludedIncluded
The org roll-upNot includedIncludedIncluded
The org security roll-upNot includedIncludedIncluded
WebhooksNot includedIncludedIncluded
Team review (more than three people stamping a field)Not includedIncludedIncluded
The succession packet (one person's graded work, assembled for a handoff)Not includedIncludedIncluded
Grading runs on your own inference on every column: your assistant CLI, your own API key or a local model. Hosted has a store cap per org: 10 GB on Starter, 25 GB on Freelancer, 50 GB on Team, 100 GB on Small Business. This demo is not a tier. It shows the paid panels unlocked on invented records, so you can see them; a Free install shows them locked.
These are not three products. The free download is the same program the paid tiers run, and it grades your imports too. The binary is free; its source is not published yet. Paid is the same binary with a key that turns on live capture, every reading level and the whole-chat read, several agents on one field, more fields and the paid panels. Hosted is that as an isolated tenant on our hosted service, set up with you. One product, four ways in.
Open source when published; binaries today Free the working tool Paid more agents, more fields Hosted set up with you

What you are buying, stated honestly

DIN grades; it does not just search. Anything can find the conversation where a decision was made. What this does is tell you whether the claim in it was ever borne out: captured at the tool-call boundary, marked receipted, logged or inferred by how it was obtained, and proved, observed, asserted or refuted by what it showed. Those two are never collapsed into one badge, because that is how an assertion ends up looking like a measurement.

The public demo at dinctrl.com is the free build's own console, running, on a field of invented records. The grading in it is not softened: refuted claims, contested decisions and two false positives of the field's own detector are all still on the screen.

A fork can rebuild the closed half. That is accepted on purpose and said out loud rather than defended against: what compounds here is the grading quality, the eval corpus, the ingestion breadth and the release cadence, and a fork freezes on the day it forks.