Free tools Windows power users keep installed
One-click scans. No signup required.
Coding agents can retain project knowledge across sessions, but persistent memory is not one universal feature. It may be a managed store attached to sessions, project instructions in files, or candidate updates extracted from old conversations. Those approaches differ in where data lives, whether an agent can write to it, and whether another CLI can use it. Current documentation describes examples from Anthropic and Gemini CLI; it does not establish a shared memory format for Claude Code, Codex, Gemini CLI, and other tools.
What “persistent memory” means for a coding CLI
Persistent memory is information that remains available after a session ends. In practice, the label covers several designs with different jobs:
As an Amazon Associate I earn from qualifying purchases.
- Context or instruction files: durable Markdown that a CLI loads into prompts, such as project conventions and build instructions.
- Managed memory stores: documents held in a service and attached to an agent session, sometimes with version history and access controls.
- Transcript-derived proposals: candidate facts or reusable skills inferred from earlier sessions and offered for review before they become active context.
These are not interchangeable. A project file can be portable if another tool is configured to read it, but it does not automatically analyze conversations or resolve conflicts. A managed store can provide stronger write controls and history, but its existence does not mean unrelated CLIs can access it.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Can Claude Code, Codex, and Gemini CLI share memory?
There is no basis in the documented sources here to say that these tools share a common persistent memory layer or automatically synchronize memories. Anthropic documents memory stores for Claude Managed Agents, while Gemini CLI documents its own context files and Auto Memory. OpenAI’s Codex repository identifies Codex CLI as a locally running coding agent, but the repository information cited here does not establish compatibility with either Anthropic’s stores or Gemini’s memory features.
Cross-agent sharing is possible only when the specific tools have a supported way to read and, where needed, write the same data. A common file such as project Markdown can be a practical interchange point if each CLI is configured to load it, but that is not the same as a shared service: the tools may interpret instructions differently, and they may not share retrieval, versioning, permissions, or conflict handling.
Before relying on cross-agent memory, verify the exact versions and integration paths for each tool. Check that both can read the same format, whether either can write to it, how concurrent edits are handled, and what access controls and history are available. No complete official compatibility matrix is established by the sources cited here.
Three documented ways to retain project knowledge
| Approach | What persists and where | Review and control | Important boundary |
|---|---|---|---|
| Anthropic Managed Agents memory stores | Text documents addressed by paths in a workspace-scoped store, mounted in the agent sandbox when attached at session creation. | Read-write is the default; read-only can be selected. Changes create immutable versions; versions can be inspected and redacted, and writes can use a content-hash precondition. | Documented for Claude Managed Agents; direct access from unrelated coding CLIs is not established. |
| Gemini CLI context files | Instructions and project context in hierarchical global, project/ancestor, and subdirectory GEMINI.md files. Configuration can include other filenames, including AGENTS.md. | Markdown files can be edited explicitly; Gemini CLI provides /memory show, /memory refresh, and /memory add to manage loaded context. |
Persistent files are not, by themselves, an automatic cross-agent memory service. |
| Gemini CLI Auto Memory | Draft memory updates and reusable Agent Skills inferred from earlier Gemini CLI transcripts; inbox items are project-local, while promoted skills can be user- or workspace-scoped. | Experimental and off by default. Candidates are reviewable and require user action to apply or promote. | Only eligible idle sessions are considered; transcript excerpts may be sent to the configured model for extraction. |
How to give a Gemini CLI project lasting context
For durable, human-editable instructions, use Gemini CLI’s context-file approach. Its documentation describes a hierarchy of global, project or ancestor, and subdirectory GEMINI.md files. The CLI combines found context files and sends them with prompts. It also documents configurable context filenames, which can include AGENTS.md; configuring a filename does not guarantee that another CLI will load or interpret it the same way.
- Put stable project guidance in the appropriate context file. Use project-level context for repository conventions and subdirectory context when guidance applies only to a portion of the codebase. Keep transient task details out of long-lived instructions.
- Inspect what the CLI has loaded. Run
/memory showto view the current context. Edit the Markdown source when something is inaccurate or stale. - Refresh after changing files. Run
/memory refreshto reload context files during the session. - Add a useful discovery deliberately. Run
/memory addto add information to the current context file when appropriate.
This approach makes the source of truth visible and editable. It does not promise that every old conversation will be searched for useful information; that is a separate capability.
What Gemini CLI Auto Memory does—and what it does not do
Gemini CLI’s Auto Memory is an experimental, opt-in mechanism for finding durable facts, preferences, workflow constraints, and recurring procedures in past Gemini CLI transcripts. According to its documentation, last updated May 13, 2026, it proposes patch files or skill drafts in a project-local inbox rather than directly changing active memory files, settings, credentials, or project GEMINI.md files.
It is not immediate recall from the current conversation. The documented eligibility conditions are that a past session has been idle for at least three hours and contains at least 10 user messages; the current session is skipped. The feature is off by default. A candidate must be reviewed and acted on before it becomes active memory or a promoted skill.
Rank #3
There is also a privacy distinction worth noticing: transcripts are local files, but Auto Memory uses model calls to analyze selected transcript content, so excerpts may be sent to the configured model. Google says the extractor is instructed to redact secrets, tokens, and credentials. That is a safeguard, not a guarantee that sensitive information can never be exposed.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →How Anthropic Managed Agents memory stores work
Anthropic describes a workspace-scoped collection of text documents optimized for Claude. A store is attached when a session is created and mounted in the agent sandbox, where the agent uses ordinary file tools. Multiple stores can be attached to a session. A read-only attachment is useful when the material is reference content that the agent should consult but not modify.
Each change creates an immutable version, giving the store an audit trail and point-in-time recovery. The documentation also describes inspecting and redacting versions, and using a content hash as a write precondition to help avoid overwriting a newer version. On self-hosted sandboxes, a worker maintains a local copy and synchronizes it; Anthropic documents a default 15-second sync interval. These are controls within the documented Managed Agents system, not evidence that another vendor’s CLI can mount the store.
Where memory is stored, and what that means for trust
Storage location depends on the design. Gemini context files are ordinary files in a hierarchy of locations; Gemini Auto Memory keeps proposed items in a project-local inbox. Anthropic Managed Agents memory is a workspace-scoped store mounted into the agent sandbox, with a local synchronized copy for self-hosted sandboxes. A local transcript or project file should not be assumed to stay local throughout every processing step: in Auto Memory, selected transcript excerpts can be sent to the configured model.
Writing memory creates a persistent trust boundary. Anthropic warns that prompt injection in untrusted prompts or tool output can cause an agent with write access to store malicious content. Later sessions may then treat that content as trusted memory. Where agents need reference material but do not need to update it, read-only access reduces this write risk. For any writable memory, treat proposed changes as code-review-worthy input: check who or what supplied the fact, remove secrets, and reject instructions that attempt to override safety or project policy.
Recommended Free Tools
Limits and maintenance to account for
Limits are implementation details, not performance measurements, and can change. Anthropic’s documentation lists the following Managed Agents limits for 2026:
Best Value
- Maximum 100 kB per memory, approximately 25,000 tokens.
- Up to 10,000 memories per store.
- Up to 8 memory stores per session.
- Version history may be deleted after 30 days, while recent versions of a live memory are retained.
Gemini Auto Memory has a different operational constraint: it is experimental, off by default, and only considers eligible past sessions under its documented idle-time and message-count conditions. Its behavior and limits should be checked against the current documentation before depending on it in a workflow.
None of the cited documentation provides comparative measurements of retrieval quality, guarantees that the right memory will be recalled, or a general solution to stale and contradictory notes. Keep durable memory concise and review it as a maintained project artifact: correct outdated facts, consolidate duplicates, and avoid treating old observations as permanent truths.
How to choose a memory design
Choose based on the scope of the knowledge and the controls you need, rather than the word “memory” in a feature name.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteQuick Recap
- Choose explicit context files for stable project conventions that should be visible and edited by people. Confirm that each CLI you use is configured to load the intended file.
- Choose a managed store when the supported agent environment needs session-attached documents, controlled read/write access, and version history. Confirm vendor and sandbox boundaries before assuming portability.
- Consider transcript-derived proposals when discovering recurring knowledge automatically is valuable and the product provides a human review step. Check eligibility, opt-in state, and whether source excerpts leave the local machine.
- For multiple tools, test the integration itself. Establish shared format, read/write behavior, permissions, conflict handling, and version support for the exact versions in use. If those are not documented, treat sharing as unverified.
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.
