Works with the AI Agents you already use
agent dreams
Persistent memory and cognitive sleep cycle for AI agents using SQLite FTS5 and Ebbinghaus decay.
$9
agent dreams
Example session with this skill installed
I'm switching to Tailwind for this project. Remember that we never use utility classes for margins; use the spacing tokens in the config. Record this as a collective rule.
- Read your context and instructions
- Compiled the agent dreams
I have persisted this architectural decision to the collective memory. It will now be available across all sessions and agents when searching for styling or Tailwind-related constraints.
Memory Consolidation Report
| Field | Value |
|---|---|
| Tool Used | dream_learn |
| Memory ID | mem_8f29ac10 |
| Scope | collective |
| Category | architecture / styling |
| Content | Use Tailwind spacing tokens for all margins. Avoid direct utility classes. |
| Decay Initialized | Ebbinghaus Curve (Strength: 1.0) |
| Search Keywords | tailwind, margin, spacing, config, styling |
Next steps
- Run
dream_recallwith the query "Tailwind margin" in your next session to verify the constraint. - Use
dream_learnto record the specific spacing token map once thetailwind.config.jsis finalized.
Connects securely to your tools. The creator never sees your data.
About this skill
LLM context windows are volatile and reset every session, forcing you to repeat instructions or re-explain architectural decisions. Agent Dreams provides a persistent cognitive memory layer and sleep cycle architecture that prevents information loss between agent restarts.
What it does
- Persistent storage saves atomic architectural decisions, pitfalls, and user preferences to a local SQLite database.
- Scope management allows memories to be shared globally across all agents or locked to specific agent instances.
- Smart filtering automatically masks API keys and sensitive credentials before data is written to disk.
- Temporal decay applies Ebbinghaus-inspired forgetting curves to deprioritize stale or low-utility information.
- Hybrid retrieval uses FTS5 keyword and semantic search to pull relevant context without bloating the token window.
How it works
- Observation triggers the
dream_learntool when a significant logic change or project rule is established. - Consolidation occurs as the system filters noise and indexes the new memory with appropriate metadata.
- Retrieval happens via
dream_recallwhen the agent needs historical context or previous solutions for the current task. - Reinforcement updates the memory's internal score every time it is successfully recalled, protecting it from decay.
Frameworks & tools
Built with SQLite FTS5 for search and Bun for the runtime. Compatible with any MCP-enabled agent host including Claude Desktop, Cursor, Windsurf, and OpenCode.
Why this beats prompting it yourself
Manual prompts and custom instructions are static and limited by the context window. This skill provides a dynamic, searchable long-term memory that evolves with your codebase and forgets irrelevant details automatically.
Use cases
- Architectural tracking to remember why a specific pattern or library was chosen over another.
- Preference persistence for keeping track of your specific naming conventions and indentation styles across sessions.
- Error avoidance by logging past bugs and their fixes to prevent regressions in future tasks.
Known limitations
Requires Bun runtime for execution. Memory decay is automatic and based on frequency of use.
Requires Bun runtime installed on the host machine. Automatic memory decay may deprioritize rarely used but important information if not recalled periodically.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 9 days ago
Permissions
No special permissions declared or detected