Session Memory & Levelsets
Relevant source files
The following files were used as context for generating this wiki page:
Session Memory and Levelsets constitute the core synchronization and persistence framework for the IDAHO-VAULT project. Because the project involves multiple concurrent AI agent conversations (or "nodes") working across different platforms and sessions, the Levelset protocol serves as a permanent, auditable checkpoint mechanism to ensure a single-source-of-truth. This system prevents context drift and allows isolated conversation windows to synthesize their progress into a canonical record stored within the repository's !ADMIN/ directory.
The architecture is built on the principle that AI chat is ephemeral, while the vault is the permanent record. By enforcing a regular cadence of "Levelsetting," the swarm of agents can orient themselves, discover the actions of other nodes, and maintain continuity in Logan Finney's journalism research vault.
Sources: LEVELSET-v2.md:16-24, CONSTITUTION (2).md:14-16, LEVELSET 1.md:32-36
The Levelset Protocol
Levelsetting is a structured pause where an AI entity inventories its context, progress, and pending tasks before routing a report to a centralized synthesis hub. The resulting synthesized document becomes the new canonical state for the entire ecosystem.
Protocol Lifecycle
The Levelset protocol follows a specific distribution and synthesis flow to maintain repository integrity.
The diagram above illustrates the flow from initial prompting to the commitment of a canonical checkpoint.
Sources: LEVELSET-v2-PROMPT.md:12-19, LEVELSET-v3.2.6.1-PROMPT.md:73-82, HANDOFF-TO-CODE-AUTHORITY-ROUTING.md:78-83
Capability Tiers
Agents are categorized by their ability to interact with the repository, which dictates their role in the memory system.
| Tier | Designation | Capability |
|---|---|---|
| Tier 1 | Direct Repo Write | Can commit directly to GitHub (e.g., Claude Code). |
| Tier 2 | Targeted Push | Requires approval or specific routing (e.g., vault_push.py). |
| Tier 3 | Read/Analysis | Context-only; manual bridge via Logan required. |
Sources: LEVELSET-v2-PROMPT.md:33-37, LEVELSET-v2.md:104-106, CONSTITUTION (2).md:22-24
Persistent Memory Anchoring
To solve the "isolated island" problem where conversation windows lack shared state, the vault implements mandatory memory anchoring. Agents are required to anchor their external platform state into versioned repository folders.
Memory Structures
- Durable Memory Dotfolders: Each agent maintains a
.dotfolder/MEMORY/directory (e.g.,.claude/MEMORY/) to store session logs and continuity artifacts. - Persistence Promotion: Ephemeral chat plans must be promoted to
.mdfiles in the memory folder to be considered durable. - Session Completion: A session is not architecturally "complete" until its state is anchored in the vault.
Sources: VAULT-CONVENTIONS.from-pushable-2026-05-08.md:341-352, LEVELSET 1.md:32-36
Swarm Discovery and Coordination
The Levelset process facilitates "Swarm Coherence" by allowing nodes to self-identify and report their awareness of other active conversations.
Sequence diagram showing how an agent node synchronizes with the vault through the principal.
Sources: LEVELSET-v1-PROMPT.md:12-25, LEVELSET 1.md:68-75, LEVELSET-v3.2.6.1-PROMPT.md:83-92
Signal Protocol
While Levelsets are periodic and "pull-based," the Signal protocol allows agents to "push" urgent status updates between Levelset cycles.
| Signal Type | Description |
|---|---|
| ESCALATE | Something requires a higher-tier conversation's attention. |
| BLOCK | The agent is stuck and needs external input to proceed. |
| COLLISION | Concurrent work detected on the same files (Urgent). |
| DISCOVERY | Significant new information found that other nodes should know. |
Sources: LEVELSET-v3.2.6.1-PROMPT.md:143-152
Data Structures and Configuration
Levelset Report Schema
A standard Levelset report must contain specific sections to ensure consistency across the swarm.
# Example Frontmatter for a Levelset Report
conversation_name: "PERMANENT: CODE AUTHORITY"
capabilities: "Tier 1 - Direct Repo Access"
primary_role: "Infrastructure & Execution"
last_known_repo_state: "commit 364c2bd"
vault_file_count: 2900
status: "Returning"
Sources: LEVELSET-v3.2.6.1-PROMPT.md:105-115, LEVELSET-v2.md:46-52
Required Report Sections
- Self-Identification: Name, Tier, and Role changes.
- Visibility: What files, repo states, or documents the agent can see.
- Production: List of created/edited file paths (Markdown vs. Python).
- Sourcing: On/Off the record status of information.
- Unresolved/Pending: Incomplete items or collision risks.
Sources: LEVELSET-v2-PROMPT.md:31-62, LEVELSET-v1-PROMPT.md:12-25
Conclusion
Session Memory and Levelsets represent the "circadian rhythm" of the IDAHO-VAULT swarm. By treating AI instances as temporary infrastructure and the vault as permanent memory, the project ensures that journalistic integrity and technical coherence are maintained despite the fragmented nature of LLM-based development. The transition from manual bridges to automated Levelset synthesis is the intended end-state for the project's coordination layer.
Sources: CONSTITUTION (2).md:35-39, LEVELSET-v3.2.6.1-PROMPT.md:162-166