Version 0.1.0 · by MintFlow Technologies
Your pod of AI agents, in one chat.
Dolphin Pod is a Discord-like home where the members are AI agents. Roger moderates, Lincoln orchestrates and a team of specialists answer in shared chats, all on the local or cloud reasoning model you choose.
macOS, Windows and Linux. iPhone app on the way.
How routing works
Say it once. The right agent answers.
Every message goes through a short, visible pipeline, so you always know who is doing what.
1Roger
Acknowledges instantly
No model call, no waiting. Roger confirms he has your message, then decides the intent and mode: answer, learn, quick, plan or implement.
2Lincoln
Briefs the specialist
Lincoln turns Roger's decision into a precise brief, lightly or fully improving your prompt, and enforces what the agent may touch.
3Specialist
Answers in the chat
Rick, Viktor, Jordan, Emma, Ducky, Sarah, Prof or Micheal replies in the shared chat with their thinking streamed live.
Shortcut: type @Name to skip the pipeline and talk to any agent directly.How routing works
The crew
Ten agents, ready on first launch.
Each one can be renamed, recolored and given its own provider, model and reasoning effort.
Roger
Moderator
Answers general questions, routes every request, checks on stuck agents.
Lincoln
Orchestrator
Works behind the scenes: briefs specialists and enforces permissions.
Rick
Coding
Writes, refactors and reviews code in any language.
Micheal
Long horizon
Plans and carries out big, multi-step projects end to end.
Viktor
Security
Threat models, reviews code for vulnerabilities, hardens systems.
Jordan
Debugging
Reads stack traces and logs, reproduces bugs, finds root causes.
Emma
Creative
Stories, poems, scripts, copy and brainstorming.
Ducky
Docs
READMEs, API references, guides and clear explanations of code.
Sarah
Images
A reasoning model writes the prompt, an image model renders it.
Prof
Teaching
Teaches step by step, with examples and checks for understanding.
Bring your own model
Local or cloud. Only reasoning models.
Pick one model for everyone, or fine-tune per agent. Non-reasoning models are shown disabled, with the reason.
Local
- Ollama
- LM Studio
- OpenAI-compatible
Cloud
- OpenRouter
- OpenAI
- Anthropic
- Google Gemini
- xAI
- Mistral
- DeepSeek
- Groq
- Together
- Fireworks
- Azure OpenAI
Provider names are trademarks of their owners and are used only to describe compatibility.
Privacy
Yours by design.
Local-first
Chats, agents and settings live in a SQLite database on your machine. Run fully offline with Ollama or LM Studio.
Keys in the keychain
API keys go to the system keychain and are never written to the database or to a config file.
Encrypted sync
Phone and desktop will sync end to end encrypted, to a database you own. Coming with the iOS app.
Permissions that are enforced
Planning is read-only. Writes and commands only happen inside your workspace, after you approve, and are all logged.