What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
To give Google Antigravity useful project context in later conversations, save stable instructions in persistent rules and keep a concise project-context document in your workspace. Google documents these files as a way to shape agent behavior and provide codebase context—not as a guarantee that Antigravity will recall every detail from previous chats. The setup below is the approach I use to avoid re-explaining the same project facts each session.
What Antigravity can carry between sessions
Antigravity discovers rules from files and injects them into agent context. Google describes rules as persistent instructions, coding standards, and architectural constraints; its Codelab characterizes activated workspace rules as always-active instructions seen in conversations in that workspace. That makes rules a practical home for durable project facts and preferences, rather than a transcript archive. Google’s Rules documentation and the Antigravity Codelab explain the distinction.
The official documentation cited here does not promise complete recall of prior conversation transcripts. Antigravity also lists Knowledge as a core capability, and Settings refers to Knowledge Items in local app data, but those descriptions do not establish that Knowledge will preserve or retrieve every past discussion. Treat persistent rules, Knowledge Items, chat history, and third-party memory tools as separate mechanisms unless documentation for your specific setup explains otherwise. Agent Capabilities Overview · Settings.
Free tools Windows power users keep installed
One-click scans. No signup required.
Choose global or workspace rules
Use a global rule for preferences that should apply across your projects, such as preferred communication style or general coding practices. Use a workspace rule for facts tied to one codebase: its architecture, stack, directory layout, conventions, and current status. Google says rule scopes accumulate, with more specific directory rules taking precedence when instructions conflict.
#1 Best Overall
| Scope | Applies to | Good fit | Where it lives |
|---|---|---|---|
| Global | Workstation-wide guidance across projects | Personal preferences and general working conventions | ~/.gemini/AGENTS.md, ~/.gemini/GEMINI.md, or modular rules in ~/.gemini/config/rules/*.md |
| Workspace | A project or workspace | Project-specific architecture, conventions, and state | AGENTS.md, GEMINI.md, or .agents/rules/*.md in the workspace |
| Directory | A relevant subtree within a workspace | Local instructions that apply only to particular files or components | Supported rule files under the relevant directory, including .agents/rules/ |
Google also documents legacy .agent/rules/ support in the described locations. Within workspace subdirectories, Antigravity walks up the directory tree from the file being read or edited and loads relevant rules at each level. Check the current Rules documentation if your repository depends on legacy paths or nested rules.
Add a workspace rule in Antigravity 2.0
- Open Customizations from the application menu or project settings.
- Select Rules.
- Choose + Workspace to create project-specific guidance, or + Global for workstation-wide guidance.
- Write concise, durable instructions and save them. For project context, include a pointer to the fuller context document if you keep one in the repository.
The documented file locations offer an alternative when you prefer to edit the workspace files directly. The exact UI labels and locations above are documented for Antigravity 2.0; consult Google’s Rules page for current details.
Rank #2
Write context that stays useful
A rule should help the agent orient itself, not become a dump of every chat or implementation detail. Google’s Codelab demonstrates a project-context rule for an unfamiliar or poorly documented repository, recording technologies, file organization, and what has been built. A practical context document can extend that example with the following stable facts:
- Purpose: what the project does and who it serves.
- Architecture and stack: the major components and technologies, with links or paths to deeper documentation.
- Key directories: where important code, tests, configuration, and documentation live.
- Current state: a brief, maintained description of what is in progress or already implemented.
- Persistent decisions: constraints or choices the agent should not casually reverse.
- Pointers: where to find detailed specifications, setup instructions, or task-specific guidance.
Keep the context short enough to scan and update it when the project changes. A stale status or superseded decision can mislead later work more than a missing detail. Avoid storing secrets or sensitive information in rules or repository files.
Keep rules, skills, and workflows in their lanes
Antigravity’s Codelab distinguishes three mechanisms: rules provide always-active context, skills supply task-specific knowledge when relevant, and workflows are reusable prompts invoked with slash commands. Put stable project facts in rules; reserve detailed reference material for skills and repeatable procedures for workflows. This keeps baseline context focused instead of loading every specialized instruction into every task. See the Codelab’s explanation.
Mind the documented size limits
Google documents a 24 KB limit per rule file and a 20,000-token aggregate budget for active global and always-on rules. If the active-rule budget is exceeded, large files may be demoted to pointers. These are configuration limits, not measures of memory quality or evidence that a setup will prevent forgetting. Keep rules focused and point to deeper material rather than placing an entire project manual in an always-active file. Rules documentation.
Rank #4
What about third-party persistent-memory plugins?
A community forum post by Shankarsan Sahoo dated August 13, 2026 describes a plugin that indexes transcripts and user preferences for recall across projects. That is the author’s description, not a Google recommendation or independent validation of reliability, completeness, safety, data handling, or compatibility. The post’s instructions may change, so check the original thread for current details before considering it: Persistent Memory Plugin forum post.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteQuick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.

