ORGANIZATION-LEVEL AGENTIC MEMORY SYSTEM

Agents forget across sessions.
HiveContext gives your team a collective brain.

A governed, multi-tenant memory layer backed by CockroachDB pgvector and AWS Lambda FastMCP compute. Capture the decisions your agents make, review them with your team, and retrieve approved context on the next task.

Explore Live Interactive Loop Browse Repositories

1. Short-Term Memory Local RAM

Working conversation scratchpad, prompt AST diffs, and immediate session buffer.

2. Semantic Memory Local Vector Index

Codebase token definitions, language semantics, and local file grep embeddings.

3. Episodic Memory Agent Session DB

Historical timeline of what a single developer's agent worked on in past branch tasks.

                  4. Procedural Memory (HiveContext)
                  CockroachDB + FastMCP
                
                  The collective intelligence layer: Governed ADRs, conventions, infrastructure rules, and post-mortems shared across all 10+ agents on the team.

01 / THE AGENT MEMORY TAXONOMY

Local Agent Memory vs. HiveContext Procedural Memory

Agent memory can be described as 4 distinct parts based on their roles: Short-Term, Semantic, Episodic, and Procedural. While individual agents manage local working state and session history, HiveContext powers the collective intelligence layer—sharing approved procedural intelligence across all agents so entire engineering teams operate as a unified, smarter brain.

Memory Dimension	Storage Layer	Scope	Purpose & Engineering Role
Procedural Memory (HiveContext)	CockroachDB + FastMCP	Multi-Agent / Entire 10+ Team	"How does our team write code, handle incidents, structure APIs, and make architectural tradeoffs?"

Isolated AI Agents (No Collective Memory)

⚠️ Mid-Sized Product Team: Linear Token Multipliers

When a team of 10 engineers each prompt frontier LLMs across separate sessions without a shared memory layer:

Token Burn Equation:
(X tokens exploring ADRs & rules + Y tokens hitting known pitfalls) × N agents × S sessions = Massive redundant token spend

Redundant Exploration
Repeated Failure Cycles
Architectural Drift

HiveContext Procedural Intelligence

Reusable Context & Strict Consistency

A single governed memory ledger stores trusted ADRs, conventions, infrastructure specs, and post-mortems:

Procedural ROI Equation:
1 validated entry in CockroachDB → Single-turn RAG retrieval (< 200 tokens) across all 10+ agents = Linear cost reduction

Guaranteed Consistency
Zero Recurring Mistakes
Dramatically Lower Token Bills

02 / PRODUCT GUARANTEES

Knowledge your team can trust.

Memory Scope

Global / Project

Share standards broadly or constrain them to a project.

Review State

Pending → Approved

Human review controls what agents can retrieve.

Conflict Signals

Semantic

Surface potentially competing guidance for review.

Memory Lifecycle

Soft Delete

Retain an auditable path before scheduled cleanup.

03 / INTERACTIVE PLAYGROUND

Simulate the closed-loop memory lifecycle.

Test how an AI coding assistant proposes a new architectural convention, triggers semantic conflict analysis, passes through the dashboard approval gate, and gets recalled in future sessions.

1. Ingest

2. Conflict Check

3. Human Review

4. Agent Recall

[HIVECONTEXT READY] Submit the proposal on the left to simulate live FastMCP ingestion, CockroachDB vector embedding, conflict detection, and agent retrieval.

04 / DISTRIBUTED SYSTEM OF RECORD

End-to-End Autonomous Architecture

Decoupled serverless compute combined with globally resilient vector transactions.

CLIENT LAYER

AI Coding Agents

Antigravity CLI, Claude Desktop, Cursor, and VS Code communicating over MCP JSON-RPC 2.0.

COMPUTE LAYER

AWS Lambda FastMCP

Serverless Python 3.12 SSE streaming endpoint managing tool execution, embeddings, and authorization.

STORAGE LAYER

CockroachDB pgvector

Globally distributed HNSW cosine vector index, multi-tenant spaces, and transactional ACID memory ledger.

GOVERNANCE LAYER

Next.js Control Panel

Real-time review dashboard for team leads to approve, reject, edit, or isolate agent-proposed memories.

Strict Scoping (Global / Project)

Semantic Conflict Alerts

Automated 30-Day Recycle Purge

Zero Cloud Infrastructure Drift

Stateless Response Streaming

05 / OPEN-SOURCE REPOSITORIES

Modular Component Repositories

HiveContext is organized into focused, independent open-source repositories designed for high maintainability and simple deployment.

MODULE 01 Next.js 16

HiveContext-Dashboard

The central control panel and human-in-the-loop review console. Provides real-time approval queues, scope isolation filters, space provisioning, and memory analytics.

Next.js 16 React 19 TailwindCSS v4 NextAuth.js CockroachDB pg

View Dashboard Repository →

MODULE 02 AWS Lambda FastMCP

HiveContext-MCP

The serverless Model Context Protocol (MCP) server deployed via AWS SAM. Exposes semantic search, memory persistence, and scheduled memory lifecycle functions over HTTP SSE.

Python 3.12 FastMCP AWS SAM Google Gemini Embeddings pgvector

View MCP Server Repository →

06 / AGENT INTEGRATION

Connect Any MCP-Compatible Client

Add the HiveContext server configuration to your AI coding assistant configuration file.

Antigravity / Claude Desktop Configuration (~/.mcp.json)

{
  "mcpServers": {
    "hivecontext": {
      "serverUrl": "https://<your-lambda-url>.lambda-url.<region>.on.aws/sse",
      "headers": {
        "Authorization": "Bearer <your-mcp-secret-token>",
        "X-HiveContext-Tenant": "default"
      }
    }
  }
}

Agents forget across sessions. HiveContext gives your team a collective brain.