← Back to Bulkhead Forge

Bulkhead Forge — product / Actions

Meetings go in. Tracked action items come out.

Actions listens where your team already decides things — Google Meet recordings, a Slack message, a voice note in the console — and turns every decision into an assigned, deduplicated action item. A tracker that fills itself.

Working MVP · Runs on our own Kubernetes cluster

Illustrative pipeline output — the stages are real, the file and counts are placeholders, not a measured client run.

The problem

Meetings end. Follow-ups evaporate.

Decisions get made in calls, then live in one attendee's notes — if anywhere. Trackers only know what someone bothered to type in, so they drift from reality within a sprint. Actions closes the gap at the source: the conversation itself.

How the same three artefacts fare today versus with Actions in the loop.
Artefact Today With Actions
Notes One attendee's doc, unread after Friday Structured analysis saved with the meeting
Follow-ups Re-typed into a tracker by hand — or not at all Extracted, assigned and deduplicated automatically
Decisions Folklore within two weeks Logged with the meeting, the owner and the date

How it works

From spoken word to tracked item

  • Phase 01

    Capture

    New Google Meet recordings are picked up automatically — across the organization, or from one member's own mailbox once they switch it on. A Drive file, an upload, a Slack message or a voice note in the console start the same path.

    Meet · Drive · upload · Slack · voice

  • Phase 02

    Transcribe

    Recordings are staged to an EU Cloud Storage bucket and transcribed by Speechmatics, with each speaker matched to an enrolled voice. Short voice notes go to Google Speech-to-Text (Chirp 3) in the EU. The transcript stays with the meeting, next to a recording you can play back.

    Cloud Storage · Speechmatics · Chirp 3

  • Phase 03

    Analyse

    A model reads the transcript under an enforced JSON schema: summary, key topics, decisions, action items, open questions, risks, follow-ups — written in the language your organization picks. The transcript itself is never translated.

    Structured output · en / ru / pl / auto

  • Phase 04

    Track

    Items land on the board with a readable key, an owner, a type, an area and components. A second pass catches duplicates — a re-raised topic gets a "discussed again" comment, not a clone.

    Dedup · high-confidence verdicts only

What you get

Tracker ergonomics, minus the typing

  • Auto-extracted action items

    Owner, type, status, due date — pulled from speech, then editable like any task you'd have typed.

  • Duplicate guard

    Every dedup verdict carries high / medium / low confidence, and only a high one merges. Anything less creates the item — a missed duplicate costs less than a lost task.

  • Decision log

    Every decision is kept with its source meeting — the "why" survives the quarter.

  • People, not strings

    Speakers resolve to real people by their enrolled voice; assignees and reporters are accounts with history, not names in a cell.

  • Board and backlog

    Readable keys such as BL-12. Filter by owner, meeting, area, component and type — the board remembers the filters you left it with.

  • Work that links up

    Sub-tasks, blocks / blocked-by between items, and descriptions and comments in Markdown — Mermaid diagrams included.

  • Wired to your repositories

    A pull request whose branch or title names an item's key is listed on it and moves it to In Review. The forge CLI works the same board with your own token.

  • Private where it should be

    An item belongs to the organization or to one person, and can be shared with people or teams. A meeting's results are visible to exactly who can see the meeting.

  • Full activity trail

    Who changed what, when — human, agent and GitHub edits alike, on every item.

Sources

A source is an adapter

The pipeline doesn't care where the words came from. Each channel is a thin adapter in front of the same transcribe → analyse → track path — so adding one doesn't touch the core.

Live today

  • Google Meet — recordings picked up automatically, organization-wide or per member
  • Google Drive and upload — any video or audio file, from Drive or your machine
  • Slack — message the bot or send a voice clip; it drafts the items and creates nothing until you confirm
  • Voice note — speak in the console, review the drafts, create them in one press
  • Manual — create and edit items by hand, in the console or with the forge CLI

On the roadmap

  • Zoom — cloud recordings from outside the Google world
  • Microsoft Teams — meeting recordings for Microsoft 365 teams
  • Email — decisions made in writing count too
  • Slack huddles — the call itself, not only the messages around it

Your team talks somewhere else? Tell us — adapters are the cheap part.

Infrastructure

Runs on our own cluster

On our cluster

  • Services and pipeline jobs on k3s, on a server we operate
  • Deployed from Git by Argo CD — a merge is the deploy
  • PostgreSQL (CloudNativePG), Vault for secrets, NATS JetStream between services
  • Keycloak for sign-in, OpenFGA for who may see and change what
  • Langfuse — prompts versioned, model calls traced

Services it uses

  • Google OAuth · Drive & Meet APIs · Cloud Storage (EU)
  • Speechmatics for recordings · Google Speech-to-Text (Chirp 3, EU) for voice notes
  • Models through OpenRouter, one per role, chosen in the console
  • A Slack app and a GitHub App, installed per organization

Roadmap

Where this goes

Now

  • Meet, Drive and upload ingest, with speakers identified
  • Slack and voice intake, confirmed before anything is created
  • Analysis, dedup, decision log
  • Board, backlog, sharing, pull requests moving items

Next

  • Zoom and Microsoft Teams adapters
  • Email adapter
  • Merged pull requests close their items

Later

  • Two-way sync with Jira & Linear
  • Cross-meeting insights per person & topic
  • Public ingest API for your own sources

Early access

A tracker that fills itself. Try it before everyone else.

Working MVP, onboarding a first cohort of teams on Google Workspace. No forms — one email is the whole first step.

contact@bulkheadlogic.com