Tech • AI • Robotics
BASE vs Graphify: The AI Memory Tool You Need to Be Using
5/10TL;DR
Bass is a new open-source system for Claude Code that builds a persistent knowledge graph for rules, behaviors, tasks and project context, aiming to give each new session immediate continuity without relying on a static claude.md file.
KEY POINTS
What Bass changes
Bass addresses two common limits in Claude Code workflows: sessions that start without prior context, and a static claude.md rules file that can quickly become outdated. It replaces that fixed model with a graph-driven system that injects relevant rules and project state at session start, so the assistant begins with awareness of active work, open tasks and user preferences.
How the system is built
The platform is described as a Rust binary with three main parts: a hook pipeline, a knowledge graph and a built-in MCP server. It supports both global and workspace tiers, allowing universal settings to follow the user across projects while repository-specific rules and context remain tied to a particular codebase.
Hooks push context automatically
A central distinction is that hooks run automatically, unlike optional skills or manual lookups. Session-start hooks preload always-on rules, operator profile details, project summaries and open handoffs before any prompt is written, while prompt and pre-tool hooks can detect keywords or file paths and inject only the rule sets relevant to the current task.
Dynamic rule injection
Bass organizes instructions into knowledge domains rather than one long rule file. Always-on domains can cover broad preferences, while specialized domains activate only when certain keywords appear, such as project-specific or development-related terms. That design is intended to improve consistency while avoiding repeated injection of irrelevant guidance.
Memory without manual curation
The system is positioned against Graphify, which requires users to query and curate a graph more directly. Bass instead keeps context at the top of the working window from the start of each session, meaning memory is already available rather than fetched on demand. It also claims scoped, deduplicated injection that can reduce token usage by as much as 70 times in long-running workflows.
Tasks, handoffs and forks
Bass includes a task system and a session-closing routine centered on an end command. Running that command can create handoff documents, register unfinished tasks, update rules and capture new build ideas as forks. A handoff records where work stopped, while a fork preserves an actionable build specification that can be surfaced in future sessions.
Cross-session relay
A ping function allows work to be handed from one open session to another using session codenames. That enables users to assign a task to a separate active session without cluttering the current context window. All sessions still reference the same underlying graph, which is meant to keep coordination organized across parallel work.
Self-learning and context control
Bass updates its graph at the end of each session, adding behaviors, decisions and project changes without requiring manual rewriting. To limit context bloat, inactive projects can automatically fall out of startup context when they go cold, then reactivate if a matching file path or keyword reappears. This acts as an archive system rather than deleting old knowledge.
Dashboard and visibility
The system includes a browser-based dashboard for reviewing domain rules, project paths, graph contents and usage analytics. A dev mode can also show which domains were loaded, which tools were used and how much context remains, giving users visibility into what information is influencing the assistant in real time.
Installation and ecosystem plans
Setup can be handled with natural-language commands such as install and scaffold, with support for new or existing workspaces. Bass can also integrate with Paul, an orchestration layer that stores project phases and milestones in the same graph. The software is free, with an enterprise version planned for teams that need shared company-level data, and broader support beyond Claude Code is under consideration.
CONCLUSION
Bass represents a shift from static prompt files and manually queried memory toward automatic, graph-based continuity in AI coding workflows. Its main promise is simple: each new session starts informed by current rules, active projects and recent decisions without overwhelming the context window.
Explain this