Continuum memory
Cross-agent memory mesh integration for Continuum. Use when an agent needs to persist context, share memory with other agents, or retrieve historical decisions and facts.From its SKILL.md
npx -y skills add anarudhan/continuum --skill continuum-memoryAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
4.8 KB, as published. Nobody here has run it
Continuum Memory Skill
Integration guide for connecting any AI agent to the Continuum memory mesh.
When to Use
- Starting a new session and want to load previous context
- Making a decision that other agents should know about
- Learning a pattern or workflow that should be reusable
- Ending a session and want to summarize for future agents
- Searching for previous decisions, facts, or procedures
Core Concepts
Memory Types
| Type | Stores | Example |
|---|---|---|
| episodic | Session transcripts, decisions | "Decided OAuth2 + PKCE for auth" |
| semantic | Facts, entities, relationships | "Project: Anarudhan, Stack: Go + React" |
| procedural | Skills, workflows, patterns | "Deploy: 1) test 2) build 3) push" |
Key Operations
continuum.session_start → Begin a tracked session
continuum.memory_write → Store a memory
continuum.memory_read → Retrieve memories
continuum.memory_search → Search across all memories
continuum.session_end → End session with summary
Usage Examples
Start a Session
Before starting work, initialize a session:
→ continuum.session_start(
agent="hermes",
project="anarudhan",
task="implement-auth-flow"
)
Write Semantic Memory
When you learn something permanent about the project:
→ continuum.memory_write(
type="semantic",
content="Anarudhan uses OAuth2 + PKCE for authentication. Go backend, React frontend.",
entities=[
{name: "Anarudhan", type: "project"},
{name: "OAuth2", type: "technology"},
{name: "PKCE", type: "technology"}
]
)
Write Episodic Memory
When you make a decision during a session:
→ continuum.memory_write(
type="episodic",
content="Decided to use httpOnly cookies instead of localStorage for JWT storage",
context="PR review feedback from Codex agent"
)
Write Procedural Memory
When you establish a repeatable workflow:
→ continuum.memory_write(
type="procedural",
name="anarudhan-deploy",
steps=[
"Run npm run build (zero errors)",
"Run npm audit (zero vulnerabilities)",
"Run security scan",
"Git commit to feature branch",
"Open PR for Kimi audit"
],
context="Frontend deployment workflow for Anarudhan"
)
Search Memory
Before implementing, check what we already know:
→ continuum.memory_search(
query="authentication decisions",
types=["episodic", "semantic"],
limit=5
)
End Session
When finishing work, summarize for future agents:
→ continuum.session_end(
summary="Implemented OAuth2 flow with PKCE. httpOnly cookies for JWT. Tests passing.",
key_decisions=[
"OAuth2 + PKCE over simple JWT",
"httpOnly cookies over localStorage"
]
)
Best Practices
- Write early, write often — Don't wait until session end
- Be specific — "Use OAuth2" is better than "Fixed auth"
- Tag entities — Helps build the knowledge graph
- Summarize on end — Future agents read summaries first
- Search before deciding — Avoid re-making decisions
Integration Methods
MCP (Recommended)
Install the Continuum MCP server in your agent:
# Claude Code
/plugin marketplace add anarudhan/continuum
/plugin install continuum@continuum-memory
# Or manual: add to your agent's MCP config
{
"mcpServers": {
"continuum": {
"command": "continuum-mcp",
"args": ["--api-key", "$CONTINUUM_API_KEY"]
}
}
}
REST API
export CONTINUUM_API_KEY="your-key"
export CONTINUUM_URL="http://localhost:8080"
# Write memory
curl -X POST $CONTINUUM_URL/api/v1/memories \
-H "Authorization: Bearer $CONTINUUM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"type": "semantic",
"content": "Project uses Go 1.24 with Gin framework",
"metadata": {"agent": "claude", "project": "anarudhan"}
}'
# Search
curl "$CONTINUUM_URL/api/v1/memories/search?q=go+framework"
WebSocket (Real-Time)
const ws = new WebSocket(
'ws://localhost:8080/ws?agent=hermes&api_key=xxx'
);
ws.onmessage = (event) => {
const update = JSON.parse(event.data);
console.log('New memory from another agent:', update);
};
Troubleshooting
| Problem | Solution |
|---|---|
| "Cannot connect to Continuum" | Check docker compose ps. Ensure continuum container is healthy. |
| "Authentication failed" | Verify CONTINUUM_API_KEY matches the one in docker-compose.yml. |
| "Memory not found" | Try broader search terms. Check memory type filter. |
| "Too many memories returned" | Use limit parameter. Add more specific entities filters. |