← All Projects
Initiative — AI Life Management System
Obsidian-vault-first AI task and goal manager where each "initiative" (project/goal) owns its own agent that reads human-authored Markdown as ground truth and schedules tasks by recursive priority.
Python (FastAPI backend)React + Vite frontendCLI interfaceYAML-driven agent state (`_initiative.yaml`)Markdown-native data model (roadmap.md, tasks.md, entries.md per initiative)
⚡ Designing a priority model that feels dynamic (momentum, dea…⚡ Keeping agent state (`_initiative.yaml`) separable from user…⚡ Making nested sub-initiatives compose sensibly without combi…
Portfolio Highlights
- →Designed and built an Obsidian-native AI productivity system where per-initiative agents read human-authored Markdown, schedule tasks via a recursive priority model, and surface focus recommendations.
- →Implemented a pluggable scoring architecture (momentum, affinity, user-defined) that computes task priority from vault state without coupling agent logic to specific heuristics.
Snapshot
- Period: November 2025
- Source: `github.com/josegibson/Initiative` (private)
- Domain: AI productivity, personal knowledge management, agent systems
- Status: Active development
Stack
- Python (FastAPI backend)
- React + Vite frontend
- CLI interface
- YAML-driven agent state (`_initiative.yaml`)
- Markdown-native data model (roadmap.md, tasks.md, entries.md per initiative)
- Pluggable scoring providers (momentum, affinity, user-defined)
What I Built
- Initiative folder schema: human-authored `roadmap.md`, `tasks.md`, `entries.md` alongside a UI-managed `_initiative.yaml` control panel per goal.
- Recursive priority system where each initiative declares a `priority_to_parent` weight enabling hierarchical focus scoring across nested goals.
- Pluggable scoring modules (`user_defined_affinity`, `momentum_pressure`) that compute task urgency from state without hardcoded rules.
- FastAPI backend exposing initiative CRUD and agent communication.
- React UI and CLI for managing initiatives and inspecting agent reasoning.
- Design principle: the vault is the single source of truth — agents read and write only human-readable Markdown, never proprietary databases.
Key Decisions
- Obsidian vault as ground truth rather than a backend database, ensuring the system is durable and user-owned.
- One agent per initiative for isolation: agents have clear scope and don't need global context.
- Pluggable scoring instead of a fixed algorithm, so prioritization behavior can be tuned per vault.