A new open-source project deploys an AI counselor that guides users through 21 days of structured self-reflection using persistent memory to reflect participants' own words back to them. Built as a Hermes Agent skill using pure Python, the tool addresses a key limitation in journaling applications: the lack of longitudinal continuity across entries.
Three-Phase Structured Questioning
The AI functions as an existential-psychology counselor, asking three carefully sequenced questions nightly across three distinct phases. Days 1–7 (See) focus on observational awareness including starting point, time usage, energy, attention, body awareness, and social masks. Days 8–14 (Understand) explore underlying patterns including fear, longing, scripts, repetition, relationships, and regret. Days 15–21 (Choose) address agency including values, freedom and responsibility, finitude, authenticity, commitment, and contribution.
This progression intentionally moves from observation to comprehension to decision-making, creating a structured psychological journey rather than free-form journaling.
Persistent Memory Creates Longitudinal Self-Awareness
The distinctive feature is persistent memory across sessions. The system records answers and weaves them into reflection summaries on Days 7, 14, and 21, creating what the creator @Forlives describes as "a mirror that slowly develops." This contrasts with most journaling tools that treat each entry as isolated, lacking continuity.
By reflecting users' own past words back to them at key intervals, the tool enables participants to observe patterns and changes in their thinking over time, fostering deeper self-awareness than traditional daily journaling.
Lightweight Technical Implementation
Built with pure Python using only standard library functions, the project has no external dependencies. Question banks are stored as JSON files, making the system lightweight and easily deployable. Users clone the repository, initialize with python si.py init --lang [zh/en], then configure nightly scheduled tasks through their Hermes Agent at a set time (e.g., 22:00).
The project supports bilingual operation (Chinese/English), uses the MIT license, and has gained 147 stars and 52 forks on GitHub since its creation in June 2026.
Key Takeaways
- The AI counselor guides users through 21 days of structured questions across three phases: See (Days 1-7), Understand (Days 8-14), and Choose (Days 15-21)
- Persistent memory allows the system to reflect users' own words back on Days 7, 14, and 21, creating longitudinal self-awareness
- Built with pure Python and standard library only, the project has no external dependencies and stores question banks as JSON files
- The tool supports bilingual operation (Chinese/English) and has gained 147 GitHub stars since launch in June 2026
- The system addresses a key limitation in journaling apps by maintaining continuity across sessions rather than treating entries as isolated