What this is
Vantik is a fork of Tegon. Tegon is an
open-source issue tracker for developers, and an alternative to Jira and Linear.
RedPlanetHQ, the original maintainer, archived it in June 2025, and nobody
develops it now. Vantik does not compete with Tegon. It continues the work from
the point where Tegon stopped.
The original project has a good core data model: projects, issues, workflows,
Kanban views, list views, and triage. But the maintainers built it for human
teams. Such a team uses AI as an assistant. It does not use an agent as the
primary actor.
This fork changes that. In Vantik, an agent creates, plans, and updates work as
its default operation. The user interface is where a person reviews and audits
that work. It is not the primary interface.
Status: this fork is early, and it is a personal project. Changes can break
your installation, and some parts still show the old Tegon brand. The code
builds and runs on a local machine. Read How to start.
In July 2026 the maintainer updated the dependencies to NestJS 11, Prisma 6,
React 19, TanStack Query 5, Tiptap 3, AI SDK 7, and zod 4. In September 2026
the webapp moved from Next.js to Vite: it builds in seconds to static files,
and its image is nginx. The webapp reads its one NEXT_PUBLIC_* setting,
NEXT_PUBLIC_BACKEND_HOST, from the server at /api/v1/config. Thus a
self-hosted installation sets it when the container starts. One update is not
complete: the ESLint 9 flat config.
The automations subsystem no longer uses trigger.dev, and Actions are gone. An
integration is one plugin that the server loads from
apps/server/src/integrations/<slug>. Connecting it is the whole configuration:
it then reacts to the record changes and webhooks it declares, and the server
dispatches that work on the redis that the stack already needs.
Attribution and license
Vantik is a derivative work of
RedPlanetHQ/tegon, and the
AGPL-3.0 license applies to it. The Tegon team gets all the credit
for the original architecture, the data model, and the implementation. The
Tegon repository is archived, and this fork continues from it. The maintainer of
this fork works independently. RedPlanetHQ and Tegon do not
control this fork, and they do not endorse it.
How to start (self-hosted)
You need Docker, or podman with the compose provider. The compose stack needs no
other setup. It runs the webapp, the API server, and all the services:
PostgreSQL and Redis. The server applies the database migrations
when it starts. PostgreSQL also provides search through FTS, pg_trgm, and pgvector.
cp .env.example .env # the default values work; change the secrets for a real deployment
echo "CREDENTIAL_ENCRYPTION_KEY=$(openssl rand -base64 32)" >> .env # the one secret with no default
docker compose up -d
Open http://localhost:3000. Then sign in with any email address. The default
.env has NODE_ENV=development. In this mode, if you configure no SMTP
server, the server writes the magic login link to its log and sends no email:
docker compose logs server | grep -A5 "magic link"
For a deployment that is not on localhost, do these three steps:
- Set
NODE_ENV=production in .env. Development mode sends the session
cookie without Secure, and it writes each login code to the log.
- Set
FRONTEND_HOST and BACKEND_HOST in .env to your domain.
- Change
POSTGRES_PASSWORD.
In production mode, a login email needs SMTP. Without SMTP, sign in with a
passkey. How to self-host tells about
production mode, the images, the credential key, and upgrades.
Search works without an external service or an API key. Set EMBEDDINGS_SOURCE=local
for semantic search with the local CPU model. Leave it unset for keyword search only.
For a hosted model, set EMBEDDINGS_SOURCE=hosted, EMBEDDINGS_BASE_URL,
EMBEDDINGS_MODEL, and EMBEDDINGS_API_KEY if the endpoint needs authentication.
Hosted models receive the search text and document text. See
search configuration for the defaults and offline use.
How to connect your own agent
Settings → My account → API (the API & Agents page) makes a token for
an agent, and gives the MCP configuration for Claude Code, Codex, Cursor, and
other clients. Two agent skills then teach the agent to
use the tracker and the knowledge bank well. Install them from your server,
which gives the copy that matches it:
DISABLE_TELEMETRY=1 npx skills add https://your-vantik-host
Or install them from this repository with npx skills add cmunte132/vantik.
The same page gives optional hooks for each agent. At the start of a session,
they tell the agent which issues it has in progress. Before the agent stops, they
hold it once if one of those issues has had no update from it for 20 minutes.
Read How to connect an MCP client
for the details.
How agent work gets checked
Delegate an issue to an agent in the hosted sandbox and the work goes round a
loop before anybody sees a pull request. One agent implements the change.
Vantik runs the repository's own test, typecheck, lint and build commands
against the tree it left. A second agent — a fresh process, in the same
sandbox, with different instructions and no sight of the first one's reasoning —
reads the diff against the issue and reports what is missing, citing a file and
a line for each thing. Those go back to be fixed, and it reads the result again.
That repeats until the reviewer accepts the work or the issue's budget runs out.
A run that runs out still delivers its branch; it finishes as Needs review
rather than as a success, and the pull request says that nothing signed it off.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
