Tickets
Jira-style boards with statuses, priorities, assignees, labels. Hand a ticket to an agent and watch it pick up the work.
Now in private beta
The OS for AI‑driven work.
Tickets, docs, channels, files, and generative tools in one workspace. AI agents arrive pre-built and pre-configured for online businesses. No prompts to write. No skills to wire. No code.
Free during beta · macOS, Windows, Linux
The shift
Work is the API now.
For twenty years we built tools that helped people type faster. Tickets, docs, chat, files. Separate surfaces with separate logins, joined together by humans copying things from one tab to another.
Then the model got good enough to do the typing. Suddenly the bottleneck isn't writing. It's giving a machine a coherent picture of the work, the history, the assets, and the people involved. The tools that were optimised for human typists are the wrong shape for that job.
blayd is the tool built the other way around. The agents arrive pre-built. The skills are configured. The prompts are written. None of it is your problem.
Open the app and your agents already know how an online business runs: triaging tickets, drafting product copy, answering customer questions, scheduling channels, queuing generation jobs. People work the way they always have. Agents work alongside them, on the same objects, with the same paper trail.
What is blayd
One workspace. Six surfaces. An agent layer that touches all of them.
blayd is a desktop application backed by a self-hosted server. You run both on your own machine. Your data, your model keys, your audit log. Nothing routes through us.
blayd is
- · A desktop app you install on macOS, Windows, or Linux.
- · A unified workspace with first-class tickets, docs, channels, files, workflows, and generation tools.
- · Pre-built agents already wired into every surface. The prompts and skills are configured. You don't write any.
- · Pre-configured for online business work: support triage, product copy, content scheduling, asset generation, order ops.
- · An invite-driven, multi-user workspace with roles, teams, and per-channel privacy.
- · Yours to run locally, behind your firewall, on your own model keys.
blayd is not
- · A framework you assemble. There are no SDKs to learn, no skill packs to configure, no prompts to write.
- · A chat wrapper around an LLM. The agent is one tab; the work surfaces are six others.
- · A SaaS we host. Your workspace runs locally. There is no central tenant.
- · A glue layer over Jira / Notion / Slack. Those tools stay where they are; blayd is for teams starting fresh.
- · Locked to one model vendor. Bring OpenAI, Anthropic, Google, or a local Ollama. The tools don't care.
How it's built
Four layers.
All four, done.
Six months of integration work, written down and shipped. The surfaces, the REST core, the tool catalog the agents call, the multi-turn runtime that picks which tool to use. You build none of it.
Most teams quietly drop a quarter and three engineers on this before they ship a single AI feature. With blayd you skip it. Open the app, the agents are already wired into every surface, pre-configured for how online businesses actually run.
- 01 · Surfaces
Tickets, Documentation, Channels, Drive, Workflows, Studio.
Every entity has a stable id, a JSON-PATCHable shape, cursor pagination, and revision history where it matters (docs, tickets, generations).
- 02 · REST core
ASP.NET Core 8 + EF Core, ~80 endpoints, OpenAPI-described.
Tickets, /docs, /channels, /drive, /workflows, /generations, /users, /teams, /invites, /auth. Cursor pagination on hot lists. SignalR hub for live chat + reaction events.
- 03 · Tool catalog
A versioned manifest of every action the agent can call.
40+ tools today: open_ticket, write_doc, post_message, generate_image, schedule_routine, … Each tool wraps a single REST call and ships with an input schema the model uses for planning.
- 04 · Agent runtime
Multi-turn orchestrator that picks tools by goal. Pre-configured for online business work out of the box.
No prompts to write, no skills to wire. The agents know how to triage support, draft product copy, schedule content, generate creative, and reconcile orders on day one. Provider-agnostic: bring OpenAI, Anthropic, Google, or a local Ollama.
What you get
Everything you'd build yourself. Only it already exists.
Six surfaces wired together, every one of them addressable from the agent layer. Skip the integrations sprint.
Documentation
Notion-style pages with a TipTap editor, revisions, drafts, per-page assignees. Plain-text projections for agent reads.
Channels
Slack-style chat with threads, reactions, shareable messages. Real-time over SignalR. No refreshes, no glue.
Drive
One file browser for what you upload AND what the generation layer produces. Folders, drag-drop, previews.
Workflows
Compose triggers, AI steps, and human approvals into recurring routines. Hangfire-scheduled, same as tickets.
Studio
Image and video generation via DALL·E, Nano Banana, Seedance, Veo, Sora. Outputs land in Drive automatically.
See it in action
Designed for humans. Built for agents.
Tickets
A board your agent can actually use.
Statuses, priorities, assignees, labels: the same affordances you'd find in a Jira clone, but every column, swimlane, and ticket is callable by name. Drop a ticket on an agent, and it picks up where the human left off, with the full discussion thread intact.
Open3
Set up SSO
Refactor /docs
Pricing v2 copy
In progress2
Agent: write Q4 brief
Drag-drop in Drive
Review1
Onboarding tour
Documentation
Notion-grade docs without the wall.
TipTap-backed editor, revision history with coalescing so the audit log doesn't drown in keystrokes, draft / published states, per-page assignees. Every page exposes a plain-text projection so agents can read without parsing rich JSON.
Q4 product brief
Draft · last edited 2 minutes ago
Studio + Drive
Generation that lands in a real file browser.
DALL·E, Nano Banana, Seedance, Veo, Sora. Pick a provider, write a prompt, output lands in Drive automatically. Move it into a folder, hide it, share it. The agent can do all of the above.
Six surfaces. One agent layer.
Tickets, docs, drive, channels, workflows, generation. Each one ships with a tool the agent can call, so handing work to AI doesn't mean rewriting how your team operates.
6Surfaces
40+Agent tools
~0Lines of glue
LocalWhere your data lives
How it works
A loop, not a pipeline.
- 01
Capture
File a ticket, write a doc, drop a file, start a channel. Or hand a routine schedule to the scheduler and let it open the work for you.
- 02
Delegate
Every surface ships with a tool: read tickets, edit docs, post in channels, generate assets. Pick a task, hand it to the agent.
- 03
Ship
Review, approve, merge. blayd keeps a paper trail across docs and tickets so handoffs between teammates (or agents) never lose context.
Early access
We're letting in 200 teams this quarter.
The private beta runs through Q3 2026. We're prioritising early-stage product, design, and research teams of 3–20 people, but the form is open to anyone, and we read every response.
- macOS, Windows, and Linux builds, signed and notarised.
- Direct line to the founders. Weekly cohort calls, private channel.
- Pinning input on the roadmap. Two of the first ten features shipped were beta requests.
- Free for life on the workspace plan once we leave beta.
From the blog
What we're thinking about.
Engineering
Why we built blayd around REST and not MCP
A short essay on tool surface design, audit logs, and why the agent should see exactly what the human sees.
Read postProduct
Chat threads, sharing, and SignalR. What shipped in v0.2
The chat module just landed. Channels, threads, reactions, mark-as-read, real-time push. Plus the dirty details on why Bearer tokens beat cookies for a desktop client.
Read postNotes
How we shipped six surfaces in eight weeks
Decisions, dead ends, and the prompt-engineering tricks behind the agent tool catalog. Spoiler: one DbContext can hold a lot.
Read postStart with one workspace.
Add your team when you're ready.
blayd runs locally on your machine, talks to a small self-hosted server, and keeps your data on your own disk.