Your AI team, and everything it knows.
A company of AI employees with one store for all its work. You name the work, they draft, you send.
A dispatcher routes every task to the bot that owns it. Outbound Slack and email wait for your yes.
Illustrative floor. Scripted, no model is called.
One request, five layers.
- 01You ask in plain words
Type the job in a chat with chief_of_staff or any bot.
- 02The dispatcher plans
A plan card shows its reasons and the assigned bots, with a countdown you can stop.
- 03Desks do the work
Every task has one owner bot, and it moves across the tasks board as it goes.
- 04You give the yes
Outbound Slack and email wait on an approval card. You answer Allow or Deny.
- 05Knowledge stays
Shared knowledge lives in a Brain you edit as text, with versioned undo.
Watch a request cross the floor.
- You askOne line in a chat with chief_of_staff.
- The desks pick it upThe dispatcher hands each part to the bot that owns it.
- It waits for your yesAnything outbound stops on an approval card.
- The Brain keeps itWhat the team learned stays, with versioned undo.
Built like a company, not a chatbot.
chief_of_staff sits at the top of your team. Under it are leads and desks, each with a defined job, and the dispatcher hands every task to the bot that owns it. Each team gets its own group chat, lead first.
Walk up to a desk
Each bot has a role, a place in the team and the model you pick for it. Pick one to see the kind of work it does.
Your server, your data, your agents and your tools.
Koshra is self-hosted, built on open source and runs on free open models, and you are not locked to one model. Nothing is rented back to you.
- Your serverKoshra runs as Docker containers on a server you control, around the clock, not in someone else's cloud.
- Your dataChats, tasks, the Brain and bot memory stay in the Postgres database on that server.
- Built on open sourceKoshra is built on an open source agent platform released under the Apache 2.0 licence.
- Free open modelsBots can run on free open-weight models, such as NVIDIA Nemotron 3 Ultra and dots3-note, with no model fees. Our own Koshra team runs on them today.
- Not locked to one modelPick the model per bot and switch mid-chat, or bring your own key for OpenAI, Anthropic, Google, OpenRouter, MiniMax or any OpenAI-compatible endpoint.
- Your agents and toolsYour bots, their jobs, their MCP servers and app connections live in your Koshra, under your own Allow, Ask and Block rules.
Model calls go to the provider you pick for each bot, and that provider sees the prompts it answers. The free open models are reached through OpenRouter. How Koshra handles your data
Change the brain, keep the job.
Pick the model for each bot. Switch in the middle of a chat and the bot keeps its job and its chat, with a clear note where the switch happened.
- 01Per botSet the model for each bot on its own
- 02Switch mid-chatA "Continued with" note marks it
- 03Backup chainA backup model chain behind each bot
- 04On every replyThe model, cost in USD and time
Bots draft. You decide.
One-click permission modes per bot, and safety rules that hold in every mode.
- External messages wait for you. Outbound Slack and email go to approval first.
- CRM guard rails. Deletes and merges in the CRM are blocked, and bulk changes need approval.
- Rules per action. Each tool can be set to Allow, Ask or Block, and creating new bots or schedules needs approval.
- Stuck runs, handled. If a run stops making progress, a bar shows it with a Cancel button.
One-click permission modes per bot. Modes only tighten: the safety rules underneath still apply in every mode. The mode you pick here also drives the demo below.
Run the floor yourself.
Pick a task or type your own. The dispatcher plans it, the bots work it, and you answer the approvals. Scripted, so no model is called and nothing is sent.
Scripted simulation. No model is called and nothing is sent.Mode: Ask risky
Illustrative board. Cards and owners are examples, and tasks from the demo above land here.
Want the app itself? Click through a replica of the Koshra app: the tasks board, a plan card and an approval card.
Hand over one job first.
Ready-made workflows the team can run for you.
research_leadPre-discovery brief
A sourced one-page brief on the account and the people before a booked discovery call. Every claim is sourced or marked unknown, and a person approves it first.
Pricing: talk to us
chief_of_staffMeeting intelligence
Each client call becomes one meeting record: summary, decisions, next steps with owners, and follow-ups held for approval.
Pricing: talk to us
product_leadReverse ICP audit
Work backwards from who bought, stalled or churned to test your stated ICP before more spend goes into it.
Pricing: talk to us
revops_deskPipeline hygiene
Stage rules and required fields, then a scheduled scan that proposes fixes in small batches. Bots never merge or delete.
Pricing: talk to us- Self-hosted, inside your stackKoshra runs on your own systems, next to the tools your team already uses. Your data and your rules stay where you put them.
- On the model you choosePick the model per bot, switch mid-chat and keep a backup chain behind it.
- Your rules, enforcedAllow, Ask or Block per tool, approvals for outbound messages and guard rails on your CRM.
Book a demo or join the waitlist
Tell us what your team should hand off.
A few lines help us prepare. Or pick a time straight away.
- Stored on our own server. Nothing is shared or sold.
- No newsletter. We only use it to reply about Koshra.
Questions from the floor.
Short answers first. Not here? Ask below and it goes into your demo request.
What is Koshra?
Koshra is a company of AI employees from Xeme AI Lab. You name the work, the bots draft it, and you decide what gets sent. A chief_of_staff bot sits at the top, with leads and desks under it, and a dispatcher hands every task to the bot that owns it.
What does the name Koshra mean?
Koshra (said KOHSH-rah) comes from the Sanskrit kosha, a treasury or store. It is the place where your AI team keeps its work: tasks, meeting records and a Brain you edit.
Who is Koshra for?
Teams that want AI employees to take on defined jobs, such as a pre-discovery brief, meeting records, a reverse ICP audit or pipeline hygiene, while a person approves anything that goes out.
Can a bot send something without my approval?
No. Outbound Slack and email go to approval first, and you answer on an approval card with Allow or Deny. Permission modes can make a bot stricter, never looser.
Can a bot delete or merge CRM records?
No. Deletes and merges in the CRM are blocked, and bulk changes need approval.
Which AI models does Koshra use?
The one you choose, per bot. You can switch mid-chat and keep a backup model chain behind it, and every reply shows its model, cost and time.
Do I own my Koshra team?
Yes. Koshra is self-hosted on a server you control and built on an Apache 2.0 open source base. Your chats, tasks, Brain and bot memory stay in its database, and your bots, rules and tools are set up there too. Bots can run on free open-weight models, and you are not locked to one model: pick one per bot or bring your own key. Model calls go to the provider you choose.
How does Koshra assign tasks?
A dispatcher shows a plan card with its reasons, the assigned bots and a countdown you can stop. Each task has one owner bot that you can tell to continue, or reassign, and it moves across the tasks board as it goes.
Where does Koshra run?
In your own systems. Koshra is self-hosted, next to the tools your team already uses, and it is a responsive web app with Web Push alerts on your phone.
How much does Koshra cost?
Talk to us. Pricing is set with you on a call, around the team you need. Book a demo and we will walk through it.





