TalvikMAK4I
An open protocol for portable, governed AI context
MAK4I is Talvik's tool-neutral protocol for AI context. It defines how context is represented, governed, versioned and reused across AI tools, models and agents — so knowledge moves between systems instead of being rebuilt inside each one.
The protocol for not rebuilding what your AI already built.
The Problem
AI context doesn't travel
Every AI tool retains knowledge differently, and context rarely survives a switch between platforms. Teams end up regenerating the same artifacts — architecture summaries, workflows, decisions — again and again, because there's no shared way to carry that work forward.
The Solution
A common protocol for reusable context
MAK4I standardizes how reusable AI context is structured, stored, governed and retrieved. Rather than tying knowledge to a single assistant or platform, it defines a portable format that any compatible tool can read from and write to — and it's deployment-neutral and storage-independent by design.
Governance
Context with boundaries and history
MAK4I isn't just copying context between AI systems. It carries the boundaries and provenance that make shared context safe to reuse.
- 1
Organization
The ownership boundary for context and the principals who can act on it.
- 2
Principal
An individual identity, with its own credentials, acting within the organization.
- 3
Project access / grant
The primary context and authorization boundary — access is granted per project.
- 4
Governed context
Artifacts carry provenance and history. New knowledge supersedes old without erasing it, so AI systems retrieve the current authoritative context with its lineage intact.
Reference Implementation
What the MVP demonstrates today
A working reference implementation covers the core protocol workflow end to end. Every capability below is exercised by the current MVP.
Create context
Record a durable decision, fact or artifact with a title, content and rationale.
Retrieve current context
Return the smallest useful set of current knowledge for a task — never the whole project.
Search context
Deterministic lookup by type, tags and status.
Supersede context
Replace a current decision with a new one; the prior version is preserved, never deleted.
History & versioning
Every version in a lineage, oldest first, with semantic version numbers.
Provenance
Author, lineage and an audit trail for every create, supersede and retrieval.
Conflict handling
Competing active decisions are surfaced, never silently auto-selected.
Integrity checks
A fail-closed integrity report distinguishes data errors from genuine conflicts.
CLI access
Operators can seed and inspect project knowledge without going through an AI client.
MCP access
Connected AI clients reach the same engine through MCP.
The reference implementation has been exercised end to end against several independent AI clients and environments it was tested with — including Claude, Claude Code, Cowork and Gemini CLI — through MCP. This reflects client compatibility testing, not official integrations, endorsements or partnerships.
Reuse Discipline
Check Before Create
- 1
Need something
- 2
Search existing artifacts
- 3
Reuse
- 4
Adapt
- 5
Only create when nothing suitable exists
Artifact Categories
Three kinds of reusable knowledge
MAK4I organizes artifacts into three broad categories.
Procedural
- Deployment workflows
- Engineering standards
- Issue triage
- Automation
Semantic
- Architecture
- Schemas
- API contracts
- Domain knowledge
Episodic
- History
- Decisions
- Incident reviews
- Retrospectives
Where MAK4I Fits
MCP connects tools. MAK4I connects their context.
Agent-to-agent protocols connect agents. MAK4I connects their context and knowledge — an interoperability layer that sits alongside connectivity standards rather than competing with them.
MCP
What tools and data can the AI access? Connects AI systems to tools and data.
Agent-to-agent protocols
Which agents can work together? Connect agents to other agents.
MAK4I
What does the organization already know that these AI systems should be able to reuse? Provides a portable, governed context layer across them.
The MAK4I reference implementation uses MCP as one integration path. The protocol itself stays tool-neutral and does not depend on MCP.
Roadmap
What's shipped and what's next
- 1
Phase 0 — Foundation
- 2
Phase 1 — Protocol Specification
- 3
Phase 2 — Reference Implementation
- 4
Phase 3 — Reusable Artifacts
- 5
Phase 4 — Hosted Registry
- 6
Phase 5 — MCP Integration
Get Started
Try the MAK4I Developer Preview
The protocol and specification are public on GitHub. The hosted reference implementation is access-controlled — email us and we'll set you up with credentials.