Uniffy brings notes, diagrams, chat, calls, files, projects, calendars, and AI agents into one private workspace for your team. Free to self host. AI is opt in and follows your permissions. No trackers, no training on your data.
You already know how to @ a teammate to tag them. In Uniffy the same @ pulls in any piece of your work. Notes, files, projects, tasks, channels, agents, rooms, calendar events. The reference becomes a live chip carrying the current title, owner, status, and activity. Edit the source once and every mention everywhere catches up. No copy-paste. No broken links. No stale screenshots.
#q3-launch
12 members · product, design, eng
live
TI
Teddy Ivanova10:42
Everything for the Q3 launch - plan, project, brand kit, and the live prep room - in one place:
Roadmap Q3 2026
Note·Georgi editing
Three flagship initiatives: agent runtime hardening, public sharing, and federated calendar.
planningleadership
2m ago·TeddyOpen
Agent runtime hardening
AGT-12·Project
Egress workers, retry logic, and observability for the agent runtime fleet.
14 of 22 tasks64%
infrapriority
1h agoOpen
Brand guidelines.pdf
PDF·3.2 MB
Logo usage, type scale, and accent palette for marketing assets.
branddesign
3d agoOpen
Launch prep - Q3
Room·Live
Quick sync on the rollout plan and the marketing copy for the announcement.
4 participants
just nowOpen
GA
Georgi Asparuhov10:46
Awesome
EngineeringAgentvia Agent10:47
Happy to assist. I'll keep an eye on the deploy and ping the channel if anything spikes.
@type to mention anything in your workspace...
enter to send
Every type Uniffy supports
Every content type. All mentionable, everywhere.
#engineering
Channel·24 members
Marko: Pushed the egress worker fix, deploying to staging now.
backendinfra
8m agoOpen
TI
Teddy Ivanova
teddy@uniffy.io
Online·writingDrafting the Q3 plan
Profile
Architecture decisions log
Note
Why we chose ConnectRPC over GraphQL, and the trade-offs we accepted on schema federation.
adrbackend
5h ago·TeddyOpen
Agent runtime topology
Diagram·Excalidraw
Egress workers, queue fan-out, and the retry path for failed jobs.
infraruntime
yesterdayOpen
Triage assistant
Agent·Claude Opus 4.7
Routes inbound issues, drafts replies, and tags by domain owner.
Running
2m agoOpen
Brand guidelines.pdf
PDF·3.2 MB
Logo usage, type scale, and accent palette for marketing assets.
branddesign
3d agoOpen
Agent runtime hardening
AGT-12·Project
Egress workers, retry logic, and observability for the agent runtime fleet.
14 of 22 tasks64%
infrapriority
1h agoOpen
Wire up egress worker retries
Task·In progress
Exponential backoff plus a dead-letter queue for jobs that exceed five attempts.
Teddy·May 9
30m agoOpen
Weekly engineering review
Event·Thursday
Thu, May 7, 10:00 - 11:00 AM
Conference Room A
Agent runtime metrics review and incident retro from last week.
yesterdayOpen
War room - incident #482
Room·Live
Triage for the egress worker outage. Live transcript + recording attached.
6 participants
just nowOpen
Every surface, native
Notes, chat, files, calendar, projects, tasks, diagrams, calls. All built into Uniffy itself. All permission aware. All wired into the same mention graph. No bolt ons, no marketplace bots, no copying between tools.
Notes
Markdown · live cursors · embedded mentions
Architecture decisions log
Teddy editing · saved 12s ago · v37
TIGATI
We pickedRoadmap Q3 2026as the canonical source over a chat thread.
Trade off was discoverability vs velocity. Background read:ConnectRPC RFC.pdf
Owner of schema federation going forward? Pinging@Teddy Ivanovafor input before Thursday review.
Lock the schema once federation owner is confirmed.
Migrate the egress workers to the new contract by Friday.
Ship behind a flag, gated byFEATURE_FEDERATED_SCHEMA.
Teddy
#adr#backend#q3
Comments · 2
GA
Georgi2m ago
Happy to own federation. Rolling out a draft proposal in @AGT-12.
TI
Teddyjust now
Done. Locking the schema after review on @Q3 Planning sync.
Tasks & projects
Kanban · ownership · due dates
To do
DLQ wiring
Dashboard
In progress
Egress retries
Done
Schema lock
Worker pool
Calendar
Week view · join links · reminders
MonTueWedThuFri
10:00 standup
14:00 1:1 Teddy
10:00 review
15:00 launch
12:00 lunch
Files
Native viewer · preview · OCR
PDF
brand-kit.pdf
XLSX
budget-26.xlsx
PNG
hero.png
DOCX
retro.docx
MP4
demo.mp4
ZIP
archive.zip
Diagrams
Infinite canvas · whiteboard · multiplayer live editing
Calls
Voice · video · live transcript
TI
Teddy●
GA
Georgi
TI
Teddy!
+ 3 more
Chat
Channels · DMs · threads · @ mentions
#engineering24live
GA
Georgi10:42
Pushed the egress fix, deploying now.
TI
Teddy10:43
Watching error rate. Will follow up.
@type to mention anything...
Agents
Five providers · every tool call runs as you · custom instructions · memory
Egress retries deploying. Two tasks open. The budget file is private to its owner, and you have no access to it, so neither do I.
tool calls gated by your roleaccess: viewer
Sharing & permissions
Access per item · people and groups · no admin override anywhere
Share · Roadmap Q3 2026
urn:uniffy:content:NOTE:8f2a...c41e
Copy linkSave changes
Access mode
Only owner
Only the owner can read or write.
Invited people
Only people listed below.
Everyone in org
✓
Everyone in the org gets a baseline role.
Baseline roleViewerapplies to all 24 org members
Explicit grants · 6
+ Add people or groups
TI
Teddy Ivanova (you)
mira@uniffy.io
OWNER
GA
Georgi Asparuhov
georgi@uniffy.io
EDITOR
@engineering 12 members
Group · backend, infra, runtime
COMMENTER
TI
Teddy Ivanova
teddy@uniffy.io · guest
VIEWER
@release-approvers 5 members
Group · joins and leaves inherit access
VIEWER
IN
Ivo Nikolov
overrides org baseline access
BLOCKED
No admin override, not even the owner of the org · every grant is recordedv37 · updated 2m ago by Teddy
Universal search
Across notes, files, chats, projects, calendar · permission-aware
egress retries7 results
Roadmap Q3 2026
Note · updated 2m ago
AGT-12 / Agent runtime hardening
Project · 14 of 22 done
Wire up egress worker retries
Task · in progress · due May 9
#engineering / pushed the egress fix
Chat · 8m ago · Georgi
AI on your terms
Uniffy's AI is an agent harness for work. Agents read, write, schedule, and organize across the whole workspace, and none of it runs until your org switches it on. Choose the model that fits the task. Bring your own provider credentials, or connect a self hosted endpoint and keep every prompt on your own hardware. Whatever you pick, every tool call the agent makes is checked against the same permission system that protects the rest of your workspace. An agent only ever sees what the user asking it is allowed to see.
One carve out, stated plainly. When an agent calls a service you connected, GitHub for example, it uses that connection's own credential. Its reach there is whatever that credential can reach, and it is the same for every member who asks. The GitHub tools we ship today only read, and an admin has to turn writing on before any integration can write at all.
Supported providers:
Anthropic· ClaudeOpenAI· GPTGoogle· GeminixAI· GrokOpenRouter· one key, many modelsSelf hosted· Ollama / vLLM / OpenAI-compatible
EngineeringAgent
Agent
Claude Opus 4.7 · Anthropic · 9 tools enabled
runningaccess: viewer
TI
Teddy Ivanova11:02
What's left in the agent runtime work? Pull the latest plan and the open tasks.
Egress retries are done and deploying. Two tasks remain. DLQ wiring and the metrics dashboard are still open. I went for the budget breakdown too, but it is private to its owner and you have no access to it, so neither do I.
cited: Roadmap Q3 2026
TI
Teddy Ivanova11:03
Mark the retries task done and bin the old runtime draft.
EngineeringAgentvia Agent11:03
tasks.update_task·Wire up egress worker retries · Done
Task closed. The note is a different matter, deleting it cannot be undone.
EngineeringAgentapproval needed11:03
About to delete Agent runtime draft v1. This cannot be undone. Approve to proceed?
notes.delete_note
tool calls gated by your role · deletes need your approval in chat9 tools enabled · 4 skills active · 12 memories
Configure agent
EngineeringAgent · saved 2m ago
Edit
Model
AClaude Opus 4.7
Anthropic
fallback: Claude Sonnet 4.61M contextreasoning: high
Each agent picks its own provider and specific model. Run Claude Opus for planning, GPT-5 for code review, Gemini Flash for triage, a self hosted Llama for sensitive work. Side by side, in the same workspace, on the same data.
Bound by permissions
Agents inherit your access
Every tool call runs as the asking user, through the same permission check the app itself runs. An agent cannot read what you cannot read, and a revoke lands on it in the same instant. Anything it cannot undo stops and asks you first. A scheduled run has nobody to ask, so it runs with the permissions of the person who owns the schedule and only theirs.
Self hosted models
No lab in the loop
Don't trust the labs at all? Point Uniffy at your own Ollama or vLLM endpoint and run Llama, Mistral, Qwen, or DeepSeek on your own hardware. No outside lab touches your prompts. Nothing ever leaves your infrastructure.
Each agent is yours to shape.
Instructions, tools, skills, memory, schedules. All of it is set per agent, never globally. Build one narrow agent for meeting prep and one that writes, and neither inherits the other's reach.
Custom instructions
Personality you write
One free text block on every agent. It sets the tone, the constraints, and the house rules the agent carries into every conversation. A preview shows the fully assembled prompt, so you read exactly what the model reads.
Tools
The exact slice you allow
Tools come in groups that map to the product. Notes, files, projects, tasks, calendar, rooms, people, search, images. A meeting agent that holds calendar and rooms cannot wander into files, even for someone who owns every file in the org.
Skills
Procedures, not prompt bloat
Short markdown procedures an agent pulls in when they apply. Four ship with Uniffy, your org writes the rest, and every edit is versioned so you can compare and roll back. A thumbs down on an answer turns into a draft skill for review.
Memory
Facts that persist
You get one memory store that every agent you talk to shares, and you can read, pin and delete all of it. Your org has its own, separate. What an agent learns in a channel stays with the channel.
Automations
Work that happens without you
Put an agent on a schedule and the Monday digest writes itself. A scheduled run executes as the person who owns it, so an automation can never become a way to run a prompt as a colleague with more access than you.
Budgets and limits
Spend you can actually see
Org budgets, per member quotas, and per minute rate limits on messages and image generation. Alerts on the way up, hard stops if you want them, and usage broken down by agent, model and key.
// Most workspace agents bypass asker permissions. // They run with the workspace owner's reach, not yours. Uniffy is the only workspace where the agent is bound by your role, and your choice of model, and the exact tool set its builder allowed it.
Sovereign by default
Your data, your servers, your call. Most workspace tools rent you access to their infrastructure. Uniffy is open source. Run the same product on your own hardware in any region you choose, or let us host it for you in the EU. No feature gating between cloud and self hosted.
One license
Open source, every line
All of Uniffy ships under the Functional Source License. Read it, audit it, fork it, run it in production at any size for free. Every release becomes plain Apache 2.0 two years after we publish it. That grant is written into the license itself.
Self host
Run on your own metal
Kubernetes, your own racks, even fully isolated from the internet. The same product as our cloud, no feature gating. Free, no user limit, forever.
EU cloud
Our hosted service stays in Europe
This one is about the service we run for you. cloud.uniffy.io lives in EU regions, GDPR aligned, no transatlantic data exposure, no US Cloud Act surprises. Self host instead and the question never comes up. Your servers, your jurisdiction.
We refuse to fork the product
Most workspace tools ship two products. A polished SaaS for their cloud, and a stripped down community build for everyone else. We ship one. Cloud and self hosted run the same code, the same migrations, the same admin pages. That choice makes the admin surface more complex than what our competitors offer. You get an org admin view and a platform operator view, even if you are the same person wearing both hats. We think that is the right tradeoff. You should not have to trust a fork you cannot read.
One codebase
Cloud and self hosted are the same product
Every commit ships to both. No "Community Edition" with features stripped out, no enterprise tier, no private fork for paying customers. If it works on cloud.uniffy.io, it works on your own servers. And the other way around.
Layered config
Deployment defaults, organization overrides
If you run Uniffy yourself, you configure things once when you install. If you are a cloud tenant, you configure things per organization through your admin pages. Same flow, same code. No "if cloud" branch hiding anywhere.
Two admin surfaces
Operator and tenant are separate roles
The admin pages belong to the org owner. That is your tenant, your settings. The platform pages belong to the deployment operator. That is billing, support sessions, cross tenant health. On cloud those are different people. On self hosted you usually wear both hats, on the same two surfaces.
Privacy posture
Operators do not read tenant content
Platform admins do not bypass the permission system. When an operator needs to see tenant content to support an incident, they request a time bound, audit logged support session that the org owner sees live and can revoke at any time. Same code path on cloud and on self hosted. There is no operator backdoor we keep for ourselves.
Honest pricing
These plans price the cloud we host for you. Every tier runs the same software. You pay for hosting, storage, service levels, and support. Never features. No AI surcharge, ever. Use your own provider keys or a self hosted model.
Self hosted
Free. No plan. No user limit.
The plans below are for our hosted cloud. Running Uniffy yourself costs nothing, at any size, forever. Open source, no card to enter.
The software is free at any size. Support is what we sell. A subscription buys a human who ships Uniffy on the other end of a ticket. Priced per deployment, per year. You never need one to run in production, and GitHub issues stay open to everyone.
SSO, SCIM, guests, custom branding, retention policies, the audit log, every agent feature. Those are product features, so they are in every plan, including the free one and the self hosted install. Your instance never calls home. Every release becomes Apache 2.0 two years after we publish it.Read how licensing works.
