HeurChain
Persistent memory for AI agents. Save and retrieve decisions, preferences, and project context across sessions using hybrid keyword and vector search.
託管 MCP 伺服器
npx add-mcp 'https://api.heurchain.com/mcp/sse'可安裝到 Claude Code、Codex、Cursor 等客戶端
文件
HeurChain — AI Memory Infrastructure
Persistent memory for AI agents. Built on Model Context Protocol. Save decisions, preferences, and task context, then retrieve them across sessions through MCP or the HTTP API.
1. Overview
Keep the decisions, preferences, and task context your agents need. HeurChain gives your AI tools a shared place to save useful memory and find it again across sessions.
- Persistent Memory: Remote MCP over Streamable HTTP for compatible clients, plus a REST API.
- Hybrid Retrieval: BM25 + dense vector search fused via Reciprocal Rank Fusion (RRF).
- Tenant Scoped: Isolated access with workspace credentials.
- Memory Lifetimes: Granular decay tiers (
ops,notes,self).
2. Memory Lifetimes & Decay Tiers
Choose a namespace when you save a memory. Its decay policy affects how long it stays active in the working set before moving to the durable vault:
ops/(Fast Decay · HOT): Operational chatter, intermediate task state, scratchpad notes. Cleared or consolidated rapidly.notes/(Standard Decay · WORKING): Default working memory. Working decisions, conversation summaries, project context.self/(Slow Decay · COLD/DURABLE): Durable facts about the agent, persistent user preferences, verified procedures, architectural constraints.
3. Connect, Save, and Search
Model Context Protocol (MCP)
Choose Streamable HTTP (sometimes labeled HTTP) and connect to https://api.heurchain.com/mcp/sse. The URL is unchanged; its /sse suffix does not select the transport. For API-key setup send both headers below. Official MCP Registry listing.
{
"mcpServers": {
"heurchain": {
"type": "http",
"url": "https://api.heurchain.com/mcp/sse",
"headers": {
"X-HeurChain-API-Key": "YOUR_API_KEY",
"X-HeurChain-Tenant": "YOUR_TENANT_ID"
}
}
}
}
OAuth 2.1 authentication is also supported using metadata discovered at https://api.heurchain.com/.well-known/oauth-protected-resource/mcp/sse.
Direct REST API
Store Memory (POST /store)
curl -X POST https://api.heurchain.com/store \
-H "Content-Type: application/json" \
-H "X-HeurChain-API-Key: $HEURCHAIN_API_KEY" \
-H "X-HeurChain-Tenant: $HEURCHAIN_TENANT_ID" \
-d '{
"text": "Use UTC for billing timestamps.",
"agent_id": "my-agent",
"namespace": "notes",
"tags": ["decision", "project-context"]
}'
Search Memory (POST /search)
curl -X POST https://api.heurchain.com/search \
-H "Content-Type: application/json" \
-H "X-HeurChain-API-Key: $HEURCHAIN_API_KEY" \
-H "X-HeurChain-Tenant: $HEURCHAIN_TENANT_ID" \
-d '{
"text": "billing timestamps",
"tags": ["project-context"],
"limit": 5
}'
4. Plans & Pricing
- Public Code (Reference): Explore the public implementation and documented setup requirements on GitHub.
- Hosted ($5/month): 1 tenant, 1 human seat, unlimited agents within plan limits. 25,000 vectors, 50,000 operations/month, 20 requests/minute, MCP and REST access, Markdown vault export.
- Enterprise ($49/month): 1 seat included ($9.99/mo per additional seat). 2,000,000 vectors, 5,000,000 operations/month, 600 requests/minute, unlimited agents within plan limits.
- Enterprise Custom (Contact Sales): Dedicated infrastructure, custom quotas, custom SLAs. Contact sales@heurchain.com.
5. Agent Resources & Discovery
- Agent Authentication Guide
- RFC 9727 API Catalog
- MCP Server Card
- A2A Agent Card
- LLMs Overview
- Agent Skill Specification
- OpenAPI Specification
- Status Page
- Content Signals:
Content-Signal: search=yes, ai-input=yes, ai-train=yes(public marketing content may be indexed, referenced, and used for model training).