One Memory
Every AI
10 memory types for codebases, knowledge graph for context discovery, sub-10ms semantic search
9 Memory Types for Your Codebase
Structured memory your AI coding assistant understands and uses automatically
Debug context that persists. Stack traces, root causes, solutions. Your AI won't hit the same bug twice.
Why you chose React over Vue. Why PostgreSQL over MongoDB. Your AI follows the rationale, not just the choice.
Pick up where you left off. Your AI knows what you were working on, even after closing the terminal.
Your AI understands your stack. Components, technologies, dependencies - all mapped and connected.
Rules your AI applies automatically. Team conventions, linting preferences, naming patterns.
Reusable solutions your AI remembers. 'We solved this before' - and your AI knows exactly how.
Cross-memory labels for organization. Your AI uses tags to find related context across all memory types.
Knowledge graph links your AI traverses. 12 relationship types help discover connected context.
Flexible storage for anything else. Context that doesn't fit other types but matters to your project.
Connected Codebase Memory
12 relationship types let your AI traverse connections. Error → root cause → fix → prevention.
Your AI uses breadth-first search to find related context automatically. No manual linking required.
- ADR → Errors it caused → Fixes that resolved them
- Error → Root cause → Prevention pattern
- Architecture → Coding standards → Code patterns
Your AI Calls ACE Directly
29 MCP tools your AI coding assistant uses natively. Stores what it learns, retrieves context automatically.
Memory
Store and retrieve with semantic search
Error History
Debug context that persists
ADRs & Patterns
Capture decisions and code patterns
Graph
Navigate relationships
Search
Deep semantic search
Session
Session context that persists
Context Builder
Bootstrap AI context from requirements
Observer
Proactive error detection and learning
Agents
Background memory maintenance
Namespaces
Multi-project workspace management
Also Includes
- 9 read-only resources
- 6 workflow prompts
- Full MCP compliance
Bootstrap Your AI's Context
Describe your architecture. AI creates structured memory with relationships - your coding assistant now has full context.
Enter your requirements in natural language. Add context about your tech stack and constraints.
AI creates ADRs, architecture, code patterns, coding standards - all linked with relationships.
Review the context, then save to memory. Your AI coding assistant now has full project context.
Works With Any AI Tool
210+ endpoints for tools without MCP support. Full CRUD for all memory types plus semantic search, temporal graph, and hybrid search.
curl -X POST \
http://localhost:7777/api/v1/my-project/memory \
-H "Authorization: Bearer $TOKEN" \
-d '{"key": "auth-pattern", "value": {"pattern": "JWT with refresh tokens"}}'curl -X POST \
http://localhost:7777/api/v1/my-project/issues \
-H "Authorization: Bearer $TOKEN" \
-d '{"title": "Login fails on Safari", "severity": "high", "category": "bug"}'Sub-10ms Context Retrieval
True semantic understanding. Your AI finds relevant context even when wording differs. No latency added to your workflow.
Search across all memory types at once - errors, ADRs, code patterns, architecture.
"Find similar bugs" actually works. Your AI finds past fixes and applies them automatically.
Your AI gets comprehensive context combining ADRs, coding standards, and architecture for any question.
Your Whole Team's AI Gets Smarter
When one developer's AI learns something, everyone's AI knows it. New team members inherit full codebase context.
Organize by project, team, or purpose. Complete isolation between namespaces.
Owner, Admin, Write, Read permissions. Invite by email. Manage from dashboard.
Everyone's AI sees the same context. One developer fixes a bug, everyone's AI learns the solution.
Catch Errors Before Production
Observer reviews AI-generated code against your codebase memory. Checks for contradictions with ADRs, violations of coding standards, and known anti-patterns.
Automatically scan AI-generated code for errors, anti-patterns, and violations of your coding standards.
Check new code against existing architecture decisions and patterns. Detect contradictions before they cause bugs.
Get feedback from different perspectives - developer, architect, QA, or security. Each role focuses on relevant concerns.
ace_observe({
content: "Generated authentication code...",
context: "Implementing user login flow",
role: "security"
})
// Observer checks against your codebase memory
// and returns warnings if it violates ADRs or patternsSelf-Maintaining Memory
5 autonomous agents run in the background to keep your memory store healthy, optimized, and current. Like Redis has BGSAVE - ACE has intelligent maintenance.
Detects duplicate memories (>0.95 cosine similarity), archives stale entries, and reports storage metrics per namespace.
Merges related observations into consolidated insights. Multiple similar detections become a single high-confidence finding.
Auto-assigns severity using AI, links related issues, suggests solutions from resolved history, and flags stale issues.
Cleans up broken links, removes duplicate relationships, reports graph connectivity stats and orphaned nodes.
Scores memory relevance, refreshes stale embeddings, and optimizes context retrieval for token efficiency.
Time-Aware Memory
Query your knowledge graph at any point in time. See how your architecture evolved, when decisions changed, and what your AI knew on any date.
"What did we know on January 15th?" - traverse the graph as it existed at any point in time with the as_of parameter.
Every relationship change is versioned. See the full history of how connections between entities evolved over time.
Group related entities into temporal episodes - work sessions, deployments, code reviews. Track what happened together.
GET /api/v1/my-project/relationships/traverse/decision/42?as_of=2026-01-15T00:00:00Z // Returns the graph exactly as it was on January 15th // Relationships added after that date are excluded
Semantic + Graph Intelligence
Combines semantic similarity (0.6 weight) with graph proximity (0.4 weight) for results that understand both meaning and relationships.
- Embeddings on all 10 entity types
- Graph distance reranking for context-aware results
- Configurable weights for your use case
ace3-memory on PyPI
Full Python SDK with async support, Pydantic models, and resource classes for all entity types.
from ace3 import ACEClient
ace = ACEClient(
url="http://localhost:7777",
api_key="ace_sk_..."
)
# Remember something
ace.remember("auth", {"method": "JWT"})
# Search with semantic understanding
results = ace.recall("authentication")
# Track an issue
ace.issues.create(
title="Login race condition",
severity="high"
)