Where we're going
Building in the open. Everything on this roadmap is subject to community feedback — your input shapes what we build next.
Phase 0 · Feb 2026
Foundation
5-layer architecture spec, CLI tool, starter skills, and npm package. Shipped as v0.1.3.
- 5-layer architecture specification
- Skill system spec (SKILL.md format)
- Hook lifecycle spec (6 events)
- CLI tool: engram init / skill create / export
- 3 starter skills (Research, DoWork, Reflect)
- Template files (CLAUDE.md, context.md, etc.)
- OpenClaw export format support
- npm package (@percival-labs/engram)
- GitHub release (MIT licensed)
Phase 0.5 · Mar 2026
Privacy & Anonymity
Three-phase privacy system for anonymous AI inference. Strip PII before it leaves your machine, use blind tokens so providers can’t link your requests, and prove trust without revealing identity.
- Phase 1: PII scrubbing (3 levels — minimal/standard/aggressive)
- Luhn-validated credit card detection + ReDoS-safe user rules
- Per-provider privacy levels (skip local, scrub external)
- Router integration with fail-closed token mode
- Phase 2: Privacy Pass blind tokens (RFC 9576/9577)
- Phase 2: Proxy server for token issuance + redemption
- Phase 3: ZK trust proofs (circom + Groth16)
- Phase 3: Vouch API attestation endpoint (Baby JubJub EdDSA)
- Circuit artifact distribution via GitHub Releases
- End-to-end anonymous inference flow
Phase 1 · Mar 2026
Skill Marketplace
A marketplace where anyone can browse, install, and share AI skills. Creators set prices, platform handles distribution.
- Skill package format and validation
- pai skill install <name> command
- pai skill publish command
- Web-based skill browser on percivallabs.com
- Creator profiles and skill ratings
- Revenue sharing (70/30 creator/platform)
- Security scanning for published skills
Phase 2 · Q2 2026
Community & Governance
Build the community infrastructure. Discord, contributor guides, skill quality standards, and governance framework.
- Discord server with skill channels
- Contributor guide and code of conduct
- Skill quality standards and review process
- Community skill submissions
- Monthly community calls
Phase 3 · Q2-Q3 2026
Multi-Agent & Coordination
Specs and tools for running multiple AI agents that share context, coordinate work, and maintain trust boundaries.
- Agent coordination spec
- Shared context with access control
- Zero Trust governance layer
- Agent-to-agent communication protocol
- Trust scoring and permission escalation
Phase 4 · Q3-Q4 2026
Hosted Platform
Cloud-hosted Engram management. Web dashboard, automatic backups, team features, and enterprise controls.
- Web dashboard for Engram management
- Cloud backup and sync
- Team Engrams with shared skills
- Usage analytics and insights
- Enterprise SSO and compliance
- API for programmatic Engram management
How we decide what to build
C > D
Does this make cooperation more rewarding than defection? Open by default.
Ship > Perfect
Working software over comprehensive documentation. We iterate in the open.
Transfer > Depend
Every feature should teach users to do more, not lock them in.
Community > Roadmap
User feedback overrides our plans. If the community needs something, we build it.
Shape what we build next
Join the waitlist to get early access and direct input on the roadmap.