Skip to content
Hoard Hoard

Roadmap

This page outlines the Hoard development roadmap.

Current Version: 0.1.1

v0.1.1 — Setup UX Hardening & Remote-First Server

  • Write-ready by default — fresh installs auto-generate a server secret and enable writes out of the box
  • Auto-generated server secret persisted to ~/.hoard/server.key (0600 perms) on hoard init / hoard setup
  • hoard setup --verify checks write-readiness, not just basic connectivity
  • Structured post-setup summary with server health, secret source, token status, and client configs
  • Health endpoint — GET /health returns liveness, DB readiness, and migration status
  • Remote bind guard — non-loopback bind requires --allow-remote flag
  • hoard setup remote — configure local clients against a remote Hoard instance
  • Docker deployment stack — Dockerfile, docker-compose.yml, Caddy reverse proxy with auto-TLS
  • HOARD_DATA_DIR — relocate all runtime paths for containerized environments
  • HTTP retry helper — exponential backoff for transient network failures in CLI
  • Duplicate and conflict detection in background worker
  • Atomic config file writes via temp file + rename

v0.1.0 — Initial Release

The initial release focuses on core functionality:

  • Local file indexing
  • Agent inbox drop folder
  • Obsidian vault support
  • Browser bookmarks (Chrome, Firefox)
  • Notion export ingestion
  • Hybrid search (BM25 + vectors)
  • MCP interface for AI agents
  • CLI-first UX
  • Connector SDK for community contributions
  • Background sync and file watcher
  • Orchestrator (Beta): agents, tasks, workflows, artifacts, events, cost

Implementation Phases

Phase 1: Foundation (Complete)

  • Schema with entities + chunks + FTS triggers
  • Entity/chunk CRUD operations
  • Local files connector with chunking
  • BM25 search
  • Basic CLI
  • Token authentication

Milestone: BM25 search returns chunks

Phase 2: Security + SDK (Complete)

  • hoard.sdk package
  • ConnectorV1 interface
  • Connector loader (trusted code model)
  • Per-token + per-tool rate limits
  • Audit logging with provenance
  • Obsidian connector

Milestone: Third-party connector can be loaded

Phase 3: Vectors + MCP (Complete)

  • Embedding pipeline
  • SQLite vector backend
  • Hybrid search with prefilter
  • MCP server with all tools
  • Browser bookmarks connector

Milestone: Claude Code queries Hoard

Phase 4: Memory + Polish (Complete)

  • Memory tools
  • hoard doctor
  • Documentation
  • Notion export connector
  • pip packaging

Milestone: v1.0 beta

Phase 5: Hardening (In Progress)

  • Edge case fixes
  • Background sync / file watcher
  • Agent inbox connector
  • Orchestrator (Beta)
  • Integration guides
  • Additional export connectors

Milestone: 1.0 stable

Future Versions

v1.5: OAuth Sources

Live connectors for cloud services:

  • Notion API
  • Google Drive
  • Confluence Cloud
  • Slack

Requirements:

  • Auth broker integration
  • Token refresh handling
  • Rate limit management
  • Delta sync for efficiency

v2.0: Sandboxed Connectors

OS-level isolation for connectors:

Hoard Core Process
    │
    │ IPC (gRPC/Unix socket)
    ▼
┌─────────────────────────────┐
│ Connector Container         │
│ ├── seccomp/AppArmor       │
│ ├── Network namespace      │
│ ├── Filesystem jail        │
│ └── Connector process      │
└─────────────────────────────┘

Implementation options:

  • Linux: bubblewrap, nsjail, Docker
  • macOS: sandbox-exec
  • Cross-platform: Firecracker, gVisor

v2.0: Public HoardHub

Self-service connector publishing:

RequirementPurpose
Code signingVerify author identity
Automated scanningDetect malware
Dependency pinningPrevent supply-chain attacks
Reputation systemSurface trusted connectors
Revocation mechanismYank compromised releases

CLI vision:

Terminal window
hoard hub search notion
hoard hub install notion_export --verify-signature
hoard hub update --check-revocations

v2.0: Write Operations

Allow agents to:

  • Create memory entries (already in v1)
  • Update entity metadata
  • Trigger re-indexing

Not planned:

  • Write back to sources (that’s the source’s job)

v2.0: Web UI

Browser-based interface:

  • Search interface
  • Connector management
  • Sync status dashboard
  • Memory browser

Known Risks & Mitigations

RiskLikelihoodImpactMitigation
Malicious connectorMediumHighCode review, provenance, curated list
Connector quality issuesHighMediumTest harness, review process
Large corpus latencyMediumHighPrefilter, documented limits
Background sync reliabilityMediumMediumLock file, debounce, tests

Performance Targets

MetricTargetConstraint
TTFA (Time to First Answer)< 20 minIncludes model download
Search latency (p95)< 300msCorpus up to 50K chunks
Search latency (p95)< 500msCorpus 50K-200K chunks
Sync reliability> 95%Successful runs / total

Contributing

We welcome contributions! See the Connector SDK for building new connectors.

Priorities

  1. Connector quality — Well-tested connectors for popular sources
  2. Documentation — Clear guides and examples
  3. Bug fixes — Stability improvements
  4. Performance — Search and sync optimization

What We’re NOT Building (v1)

  • OAuth live connectors (deferred to v1.5)
  • Write operations (deferred to v2.0)
  • Web UI (deferred to v2.0)
  • Sandboxed execution (deferred to v2.0)
  • Public connector marketplace (deferred to v2.0)

Feedback

Have ideas or feature requests? Open an issue on GitHub.