Changelog
What changed in the Koshra app and on this site. Dates are in India Standard Time. Follow along with the RSS feed.
Copy correction: no overall ranking
- Home, docs, security, comparisons, llms and generated pages now mark Mac, dedicated computers, advanced Knowledge, complete Plays and OS sandboxing as coming soon
- Turnstone and OpenBot pages checked 7 November 2026 against public primary sources. Competitor features are advertised, not independently tested
- Older changelog entries below are kept as they were written
Site v7: comparisons, catalog apps, licence and residency answers
- Comparison pages for Rerun, Lindy, Dust and n8n, each dated and sourced, with where Koshra is behind and what is coming
- Pages for apps in the Koshra app catalog, such as Salesforce, Zendesk, Jira and Microsoft Teams, with the roles that use each one
- Use cases for agencies and for founders, built from the roles in our own workspace
- Direct answers on reselling Koshra under its Apache 2.0 open source base, and on keeping data in India, the EU or wherever your server is
- An RSS feed for this changelog
Create a bot from a description, 48 role templates and bot icons
- Describe the bot you need in plain words and Koshra drafts it for you to review
- Or start from one of 48 role templates
- An icon picker with 78 icons and 12 colours, and every bot in a space now has an icon
- New bots start on a free model, and you can change it per bot
An independent uptime monitor behind the status page
- The status page now reads an independent uptime monitor that checks the API, web app, website and integrations every minute
- A public status page at status.lumenlab.tech, with the app's own status feed as a fallback
- Each service shows its uptime for the last 24 hours
Site v6.3: own your AI team
- A new home page section: your server, your data, your agents and your tools, with nothing rented back to you
- Self-hosted, built on an Apache 2.0 open source base, free open-weight models and no lock-in to one model, now on the home page, Security, the comparisons, the FAQ and the Markdown copies
- Model calls still go to the provider you pick for each bot, and the site says so
A REST API and an MCP server for other agents
- Space admins make scoped API keys in Koshra; the secret is shown once and only a hash is stored
- REST endpoints for bots, the Tasks board and task status, with an OpenAPI 3.1 spec
- An MCP server with list_bots, read_tasks_board, get_task_status and create_task
- Work that needs approval never starts from a key
- A public health check and status feed, checked every 5 minutes
Site v6.2: Markdown for agents, use cases, integrations and a glossary
- Every page is also served as Markdown, at its path plus .md or with Accept: text/markdown
- llms.txt and llms-full.txt are now built from every page on the site
- Use cases for nine teams and a page for each role, from the roles in our own Koshra workspace
- An integrations directory, a glossary, five new docs pages and a page for each workflow
- A clickable replica of the Koshra app with scripted data, and new illustrated motion
- A comparison page for Reason Machines
- A live status page for the API, web app and website, read from the Koshra app's status feed
- Docs for the REST API and MCP server
Site v6.1: docs, changelog, comparisons, security
- Docs: getting started plus pages on AI employees, the dispatcher, approvals, memory, integrations, models and self-hosting
- This changelog, a security page, a status placeholder and comparison pages for Turnstone and openbot
- Colour contrast raised to WCAG AA across the site
Site v6: the product is now Koshra
- New name, Koshra, from the Sanskrit kosha, a treasury or store
- New mark, favicon set and app icons
- New line: Your AI team, and everything it knows
Site v5.1: answers for AI assistants
- llms.txt and llms-full.txt, a sitemap and structured data
- A visible FAQ
Site v5: The floor
- A new design built around an office floor of AI employees, with scroll-driven layers
- Self-hosted fonts and full reduced-motion support
Tasks board, plan card and guards
- Tasks board with one owner bot per task
- Dispatcher plan card with a countdown you can stop
- Continue and reassign, with guards on who may reassign
- Branch from any message
- Per-bot permission modes and looks
- Reply time on every message and a configurable stuck threshold
New app design
- Light, dark and system themes, with palettes and looks
- Model and USD chips on replies
- A stuck-run bar with cancel
CRM guard rails
- CRM deletes and merges blocked
- Bulk CRM changes and writes over 25 records need approval
- Tasks added by hand get the same limits and policy checks as dispatched ones
Phone alerts, Brain undo, templates
- Web Push alerts on your phone
- Brain editing with versioned undo
- Templates and writing voice
- Model picker in the chat composer
Model switch and cost per reply
- Switch a bot's model mid-chat
- Model and cost shown on every reply
- Runs waiting for input expire instead of hanging
Also as Markdown for AI agents.
Last updated