Why Kortny instead of ChatGPT or other AI tools?

Kortny acts where other AI tools only answer. It is connected to your real tools, including scoped Composio app connections and native tools for search, Slack files, memory, channel history, and PDF artifacts. Context persists when a fact is confirmed, while episodic task memory keeps follow-ups grounded in the right thread or channel. You stay in Slack, and you own the runtime.

• Bots answer. Kortny acts because it is connected to your tools.
• Confirmed memory remembers team conventions across conversations.
• Works inside Slack, with an operator dashboard for traces, costs, memory, and tools.
• Self-hosted, so messages, task events, memory, artifacts, and model routes stay under your control.

Capabilities

Research and Writing

• Web research through Brave Search when a question needs current information.
• Document drafting for specs, briefs, OKRs, and PDF artifacts from rough Slack notes.
• Async standup digests that collect updates and post the useful version back to Slack.

Engineering

• PR review that can route to an analysis model, read diffs, flag risk areas, and draft review comments.
• Retro summaries that cross-reference Notion retros with open Linear tickets.

Team Operations

• Confirmed memory for team conventions and workflow preferences, with inspect and forget flows.
• Procedural skills for research synthesis, status recaps, document iteration, and Slack replies.
• Multi-channel support, with context scoped per channel or DM.

Use Cases

Retro Summary

Ask Kortny to pull the Q3 retro from Notion and cross-reference it with Linear. It identifies open action items, finds unassigned tickets, and can create a Notion tracker linked back to the Linear ticket, all from one Slack message.

Web Research

Ask for a competitive analysis and Kortny scrapes public docs, changelogs, and marketing pages. It returns tables and summaries and saves the full research brief to Notion with cited sources.

PR Review

Paste a GitHub PR link in Slack. Kortny reads the diff, flags blocking issues, posts inline review comments directly on GitHub, and creates Linear tickets for follow-up work.

Async Standup

Kortny prompts the team at a set time, collects updates, and posts a clean digest with blockers highlighted. It can also identify who has production access to resolve blockers and send them a message.

Integrations

Two useful capabilities ship in the native layer: real-time web search via Brave Search, and Slack file reading for PDFs, Markdown, CSV, HTML, and plain text. Generated PDFs are stored as tracked task artifacts.

Third-party integrations run through Composio, providing access to 1,000+ apps. Each connection is scoped per user, per channel, or per workspace, and missing connections are treated as recoverable setup steps.

Memory

Kortny has three memory surfaces with different scopes:

• Confirmed facts: preferences and facts you explicitly confirm, with provenance back to the source task or message.
• Episodic task memory: compact records of completed, failed, or cancelled tasks with tools, artifacts, and source references.
• Procedural skills: repeatable ways of working that operators can inspect by version.

All stored memory is fully inspectable via the admin dashboard. Every entry shows who proposed it, who confirmed it, where it came from, and whether it was superseded or forgotten. Users and admins can delete active facts at any time.

Architecture

When you mention Kortny in Slack, the message moves through Slack ingress, intent detection, a durable Postgres task queue, context assembly, model routing, native or Composio tools, ordered task events, and a Slack-native final response.

Kortny is self-hosted using Docker. Docker Compose starts Postgres, migrations, Slack Socket Mode ingress, the worker, and the local operator dashboard.

The LLM router supports a default model plus optional cheap, standard, analysis, document, and high-reasoning models. A single task can use multiple LLM calls and record the provider, model, tier, tokens, and cost for each one.

Admin Dashboard

A separate operator dashboard ships with Kortny for managing the deployment. It covers three areas: Observe (overview, task detail, LLM usage, artifacts, task events, and cost analytics), Govern (memory governance, access control, and member self-service), and Configure (integrations, tool registry, model routing, and system diagnostics).

Admins can restrict Kortny from specific channels entirely, preventing it from reading, responding in, or surfacing suggestions from excluded channels.

Roadmap

Shipped

• Slack-native coworker for DMs, mentions, files, and threaded follow-ups.
• Durable task runtime with a Postgres-backed queue, retries, and failure handling.
• LLM orchestration with OpenRouter support, model tier routing, cost tracking, and token tracking.
• Native tools for web search, PDF generation, Slack file reads, and memory read/write.
• Composio integrations with a 1,000+ app catalog, OAuth, and API-key connect flows.
• Memory for confirmed facts, episodic task context, and procedural skills.
• Admin dashboard for usage analytics, traces, memory governance, and access control.
• Observability through structured logs, task events, and LLM call telemetry.

Planned

• Observe Framework for proactive workspace observation and suggestions with human review gates.
• Polished document generation with templates and brand-aware layouts.
• Semantic memory with pgvector.
• Approval gates that require human sign-off before sensitive or destructive tool calls.
• Multi-workspace OAuth install flow.
• Containerized sandbox for safe code execution.

Open Source

Kortny is MIT-licensed and fully open source at github.com/boffti/kortny. Self-host on your own server or cloud account, own your Postgres data, choose multiple model tiers, build custom tool handlers, and audit the full source. Every task has request, result, LLM usage, artifacts, and an ordered task event timeline.