Local-First. Closed Binary.

Persistent Vector Memory
for AI Agents

One file per agent. MCP-native. Local-first.The intelligence layer for agent fleets.

Pre-built binaries for macOS, Linux, and Windows. No package manager required.

Core Capabilities

Built for Multi-Agent Intelligence

Everything your AI agents need — from persistent memory and semantic search to fleet-scale operations and compliance.

Core Architecture

The foundation that makes phorvec different.

One File Per Agent

Each agent gets an isolated .avdb file — portable, composable, zero shared state. Move agents between machines by copying a single file.

Hybrid Search

Vector similarity + BM25 keyword search fused with Reciprocal Rank Fusion. Configurable alpha weighting, quoted-string auto-boost, debug output.

Knowledge Graphs

Automatic entity extraction with 8 built-in types. 5 graph tools for search, query, link, custom entities, and analytics.

Vector Quantization

Float32, Int8 (4x savings), and Binary (32x savings) quantization with automatic threshold transitions based on collection size.

Context & Memory

Complete lifecycle management for agent knowledge.

Context Branching

Create, switch, and merge context branches. Snapshots saved as portable .vlbrain bundles. Timeline replay with session filtering.

Context Decay & Optimization

Half-life decay with per-category formulas. Anchor exemption. Intelligent pruning keeps the most relevant context within token budgets.

Memory Consolidation

4-phase pipeline: dedup (>92% similar), compress old items, associate (create links), archive low-importance. Non-destructive — originals preserved.

Retention Policies

Active → Archived → PurgeQueue lifecycle zones. Configurable grace periods, sweep scheduling, and anchor exemption from all cleanup.

Security & Compliance

Enterprise-grade security built into every operation.

Audit Trails

Append-only JSONL logging with correlation IDs. Per-operation metadata: timestamp, type, duration, result status. Daily rotation, configurable retention.

Secret Detection

Pre-storage scanning for API keys, tokens, credentials, and PII. Configurable policies: reject (block) or redact (mask and store). Denied path patterns.

Air-Gapped Operation

Local ONNX embeddings (23MB model, zero API calls). All data on-device. .vlbrain bundles for offline transfer. No mandatory cloud sync.

Compliance Posture

Append-only JSONL audit trail with correlation IDs and structured error chains, PII scanning, configurable retention zones (Active → Archived → PurgeQueue), and RBAC — the building blocks regulated workloads need.

Multi-Agent & Teams

Coordinate fleets of agents with shared intelligence.

Team Memory

3 sharing modes: FULL (bidirectional), PROTECTED (read-only to team), ISOLATED (no sharing). Team-wide search, dispatch, and feed.

Conflict Detection

Automatic scanning for contradictory decisions across agents. Regex-based negation detection, severity levels, and dismissal — no LLM overhead. See the runnable demo in Docs.

Cross-Agent Insights (DRE)

Divergent Recombination Engine finds maximally distant compatible memories, blends them, and validates novelty. Creates knowledge no single agent had.

Geometric Failure Recovery

Computes repair vectors away from failure embeddings. Exclusion filtering rejects similar failures. Agents find new paths instead of retrying.

Enterprise Operations

Observe, govern, and manage agent fleets at scale.

Agent Health Monitoring

6 EMA-tracked metrics: coherence, contribution, prediction error, latency, success rate, and trend. Fleet-wide heatmap with threshold alerting.

Compute Governance

Per-agent token budgets with typed costs (retrieval=1, model_call=5, tool_exec=3). Reliability-based allocation. Denial counters for observability.

Agent Testing & QA

YAML test suites with IR metrics (precision, recall, F1). Baseline management, regression detection, and side-by-side benchmarking.

Skill Library

Cross-project knowledge persistence. Store and retrieve reusable error-fix pairs, patterns, and organizational knowledge bundles.

Data & RAG

Index, search, and retrieve from your entire codebase.

RAG Indexing

Index files and directories with language-aware chunking for Rust, Python, TypeScript, Go, Java, and more. .gitignore-aware, binary-safe.

100+ MCP Tools

Complete tool coverage: context, memory, search, teams, quantization, graphs, security, testing, governance, and utilities.

Query Cache & Analytics

Semantic query cache with configurable TTL. query_explain for search internals. query_profile for latency breakdowns. Full analytics dashboard.

Storage Management

Integrity verification, corruption repair, write buffer management, and LanceDB compaction. Configurable storage backends (LanceDB or Lite).

Distribution & Trust

Closed-source signed binary, tier-separated at build time, downloaded direct from us.

Pre-Built Binary

One executable per platform — macOS, Linux, Windows. No package manager, no Python runtime, no daemon, no external database. Drop it on your PATH and point your MCP client at it.

Two Storage Backends

LanceDB (default) for compressed Parquet-backed indexes, or Lite — pure-Rust HNSW + SQLite for constrained environments. Switch via PHORVEC_STORAGE_BACKEND.

Portable Bundles

.vlbrain bundles serialize context branches, snapshots, and skill libraries into a signed file. Move them between machines, archive them, or import them air-gapped — no rebuild, no re-embedding.

Signed Releases

Every release ships a phorvec-mcp.manifest.json signed with our Ed25519 release key. Verify the manifest, then SHA-256 each binary against it. Self-checksum at startup refuses to run if the binary has been patched.

Quick Start

Up and Running in Minutes

Install, configure, and your agents have persistent memory.

1

Download

Pre-built binary for your platform — no package manager.

step-1
2

Configure

Add to your MCP client.

step-2
3

Use

Your agents now have persistent memory.

step-3
$
See It In Action

Powerful Agent Memory in Action

Watch phorvec handle real agent workflows — from context storage to cross-agent knowledge synthesis.

Context Store

architecture / one-file-per-agent

Nexus Registry
nexus.json
code_agent
.avdb
test_agent
.avdb
review_agent
.avdb
Each .avdb file contains:
Vectors
Key-Value
Graph
Hybrid Search Flow
Query
Vector
BM25
RRF
Results
Pricing

Dual-Track, Developer-Friendly Pricing

Start free with the full core engine. Scale solo productivity with MCP Pro & Max. Move to per-project pricing for infrastructure-scale teams.

Prices shown in USD. Billed in ZAR at the current exchange rate.

MCP Dev-Tool Track — flat monthly

Community

Free

Full core engine for solo developers.

  • 100+ MCP tools
  • Local ONNX embeddings (zero API calls, works air-gapped)
  • Hybrid search — vector + BM25 + reciprocal rank fusion
  • Knowledge graphs with 8 entity types and 5 graph tools
  • Context branching, snapshots, timeline replay
  • Memory consolidation + half-life decay
  • RAG indexing for files and directories
  • Skill library — search & retrieve
  • One file per agent (portable .avdb)
  • Bug reports via GitHub Issues (no SLA)
Get Started
Most Popular

MCP Pro

$19/mo

Individual power users — testing, baselines, skill publishing.

  • Everything in Community, plus:
  • Agent testing suite — YAML scenarios with IR metrics (precision, recall, F1)
  • Baselines & regression checks
  • Side-by-side agent benchmarking
  • Skill library publishing (publish, deprecate, sync, ingest)
  • Email support (24h response)

MCP Max

$39/mo

Small teams — shared memory, conflict detection, workflows.

  • Everything in MCP Pro, plus:
  • Team memory — FULL / PROTECTED / ISOLATED sharing modes
  • Cross-agent conflict detection (regex negation patterns, severity levels)
  • Team-wide search, dispatch, and feed
  • Team workflows, personas, and consolidation
  • Email support (24h response)

Infrastructure Track — per-project

Team

$49/project/mo

A single hosted phorvec instance for your whole team.

  • Everything in MCP Max, plus:
  • Centrally hosted deployment — one phorvec instance for the whole team
  • Shared team sync — every developer's MCP client reads/writes the same memory state
  • Single license per project (no per-individual seats to manage)
  • Priority email support (next business day)
  • Quarterly product roadmap input

Enterprise

Custom

Multi-agent inference, fleet operations, and embedded deployment.

  • Everything in Team, plus:
  • Multi-agent inference:
  • Cross-agent insight discovery (DRE) — find connections between agents
  • Geometric failure recovery — agents reroute away from failure embeddings
  • Compute governance — typed token budgets and per-agent allocation
  • Fleet operations:
  • Agent health monitoring (6 EMA metrics, fleet view)
  • Fleet-wide intervention orchestration
  • Org-wide pattern sharing and retention sweeps
  • Security & compliance:
  • Append-only JSONL audit trail with correlation IDs and structured error chains
  • Secret detection (reject / redact policies)
  • RBAC + API key auth
  • Storage repair & integrity verification
  • Deployment rights:
  • Air-gapped operation with offline-verifiable license files
  • Embedded deployment (drones, vehicles, robots, edge devices)
  • OEM redistribution bundled with your hardware
  • Support:
  • SLA-backed support (1h / 4h / 24h tiers)
  • Onboarding & integration engineering
Contact Sales

Detailed Feature Comparison

FeatureCommunityMCP ProMCP MaxTeamEnterprise
Core Engine
100+ MCP tools
Vector storage + hybrid search
Knowledge graphs with entity extraction
Context branching, snapshots, timeline
Memory consolidation + half-life decay
RAG indexing (file & directory)
Vector quantization (Float32/Int8/Binary)
Local ONNX embeddings (no API calls)
Secret detection at write time
Agent Testing & Skills
Agent testing suite (YAML scenarios)
Baselines & regression checks
Agent benchmarking
Skill library (publish, sync, ingest)
Multi-Agent Collaboration
Team memory (FULL/PROTECTED/ISOLATED)
Conflict detection across agents
Cross-agent search and dispatch
Team workflows
Team personas & skills
Licensing & Deployment
Local install (per-developer binary)
Centrally hosted instance (whole team)
Shared team sync (one memory state)
Per-individual licensing
Per-project licensing (whole team covered)
Air-gapped operation (no API calls)
Offline-verifiable license files
Embedded deployment rights (OEM)
Multi-Agent Inference
Cross-agent insight discovery (DRE)
Geometric failure recovery
Compute governance (energy ledger)
Agent health & fleet monitoring
Org-wide pattern sharing
Retention management & sweeps
Security & Operations
Audit log readable via security_audit tool
RBAC + API key auth
Storage integrity verification
Storage repair (offline rebuild)
Support
Community support (GitHub Issues)
Email support (24h response)
Priority email support (next business day)
SLA-backed support (1h / 4h / 24h)
Dedicated account manager

Works With Your Stack

Phorvec works with every popular IDE and CLI that speaks MCP. First-class setup for the clients below, and any other MCP-compatible client too.

CD
Claude Desktop
CC
Claude Code
Cu
Cursor
Ws
Windsurf
Cl
Cline
VS
VS Code
Ze
Zed
Ai
Aider
Ag
Antigravity
Cx
Codex CLI
100+
MCP Tools
3
Transport Protocols
Local-first
ONNX embeddings
<5ms
Search Latency

Give Your AI Agents
Persistent Memory

Pre-built binaries. Free Community tier. No account required for the free download.