Distilly
Distills messages, documents, interviews and public sources into reusable Person Profiles packaged as Agent Skills for AI agents.
At a Glance
Distilly core is released under the MIT License, which is OSI-approved, and can be used, modified and distributed freely.
Engagement
Available On
Listed Oct 2026
About Distilly
Distilly is an open-source project by titanwings, formerly known as Colleague Skill. It turns material you provide, such as chat logs, documents, interviews and public sources, into a source-grounded Person Profile that AI agents can reuse. The current release packages each profile as an Agent Skill that supported local agent hosts can install and invoke.
What It Is
Distilly is a person-modeling layer for agents. Its creator Skill, named distilly, builds profiles from observable experience, decision patterns, expression and ways of working. The README states it does not claim to clone the person behind them. Profiles can be made for colleagues, relationships (partners, family, friends) and celebrities or public figures, including fictional characters.
How the Workflow Runs
You ask your agent to use Distilly to create a Person Profile for a person, choose a family (colleague, relationship or celebrity), add a description or source material (every field is optional), then review the result. Distilly generates a profile packaged as an Agent Skill named {character}-{slug}. A separate flow interviews you for about five minutes to build your own profile, which stays local at ~/.distilly/owner/profile.md with a short summary added to your agent's session file.
Person Families and Sources
Each family has its own collection strategy and profile structure. The colleague family builds a Work Skill plus a Persona, with Lark, DingTalk and Slack collection. The relationship family organizes expression patterns, emotional triggers, conflict patterns and repair patterns. The celebrity family ships a six-dimension research toolchain: subtitles, transcript cleanup, research merge and quality check. Supported inputs include Lark, DingTalk and Slack collection, public X posts through a third-party service, WeChat exports, PDFs, images, screenshots, email files, Markdown and pasted text.
Evolution and Host Support
Profiles can be updated by appending files, which are analyzed and merged without overwriting earlier conclusions. Conversational corrections are written to a Correction layer, and each update is archived so you can roll back versions. The README documents native local Skill discovery for Claude Code, Hermes, OpenClaw, Codex, DeepSeek Harness, Pi, Grok Build and OpenCode. Grok Bot is listed as a saved-Skill workflow preview, with direct repo installation not yet verified.
Community Discussions
Be the first to start a conversation about Distilly
Share your experience with Distilly, ask questions, or help others learn from your insights.
Pricing
Open Source (MIT)
Distilly core is released under the MIT License, which is OSI-approved, and can be used, modified and distributed freely.
- MIT License
- Python 3.9+
- Packages each Person Profile as an Agent Skill
- Third-party services (e.g. the metered Xquik service for public X posts) are optional external costs
Capabilities
Key Features
- Person Profiles packaged as Agent Skills
- Colleague, relationship and celebrity families
- Six-dimension celebrity research pipeline
- Lark, DingTalk and Slack collection
- WeChat, email, PDF, image and Markdown inputs
- Append-based profile evolution
- Conversation corrections via a Correction layer
- Version archive and rollback
- Owner profile setup interview
- Native Skill discovery across eight agent hosts