Abbey
Human-facing companion
Designed for thoughtful teaching, creative work, and technical collaboration.
A persona contract; assistant behavior depends on the configured backend.
Explore Abbey →Quesar by MLAI · Private AI operations.
Human imagination.
Adaptive intelligence.
Build assistant workflows, memory systems, and developer tools around inspectable context. Explore the architecture, follow the source, and define the evidence your project needs.
AI should amplify human creativity.
Our direction is to help people learn, build, reason, and create while keeping human agency at the center.
The MLAI ecosystem
Distinct roles. One inspectable architecture.
Abbey, Aviva, and Abi are persona roles in ABI. WDBX supplies the memory substrate. Read each implementation boundary before integrating.
Aviva
Direct expert mode
A response mode for concise, candid technical answers, concrete next actions, and explicit uncertainty.
A local persona profile; the role does not establish model quality.
Explore Aviva →Abi
Coordination layer
Evaluates intent, risk, context, and response style to select a persona or controlled blend.
Local routing and scheduler helpers; broader governance remains partial.
Explore Abi →WDBX
Memory with provenance
Local Rust storage and retrieval components preserve durable records, vectors, and causal history.
Storage integrity does not make a statement true; production multi-host deployment is not established.
Explore WDBX →- 01Your context
- 02ABI routing
- 03Selected persona
WDBXContext and storage · writes report their result
Memory with provenance
An answer should carry its context.
Explore the sources, retrieval weights, and causal history behind a memory system. This example illustrates an inspectable trace.
Answer
Keep the German customer records in-region.
Because of, strongest first
- runbook/retention-policy.md0.94
EU records stay in-region; no cross-border egress.
blk_8f21c4 parent a17d , via Aviva
- contracts/dpa-2026.pdf0.71
Processor may not sub-process without written notice.
blk_8f21b0 parent 6c02 , via Abi
- thread/eng-platform#4120.38
Earlier draft, superseded and kept for audit.
blk_8f2196 parent 3e9f , via Abbey
For builders
Start with the source.
Explore ABI’s Rust framework, local CLI, and MCP contracts. Follow the setup and verification guidance before connecting models, tools, or memory.
Local CLI example — requires ABI source setup
./tools/cargo.sh build -p abi-cli
./target/debug/abi agent plan "stage a safe WDBX refactor"Local framework and contracts. Hosted Quesar APIs and platform SDKs are not published.
Research and simulation
Explore the ideas. Inspect the evidence.
Read source-backed work on memory, retrieval, model orchestration, and provenance, with implementation status beside the research.
In development
Visit the research library →NYON: deterministic worlds
A Rust strategy game with a separate offline Galaxy Workshop sandbox.
Workshop development and desktop/browser qualification remain in progress. NYON is a related source project.
Explore NYON’s source →For organizations · Services
Choose the next engineering decision.
Define a private AI engagement around retrieval, tool permissions, deployment topology, and evaluation. Scope and acceptance criteria are agreed for each project.
Assess the system
Scope an audit of workflows, data paths, and approval gates. Use workshops and runtime profiling to turn open questions into a risk register, architecture decisions, and an optimization backlog.
- Autonomy Readiness Audit
- Executive & Engineering Workshops
- Model & Runtime Optimization
Build a bounded workflow
Scope retrieval architecture, agent roles, and deployment topology. Translate research into an API contract and test plan, with the data boundary and tool permissions defined before implementation.
- WDBX Retrieval Architecture
- Multi-Agent Orchestration
- Private AI Deployment
- Research Translation
Harden the release path
Scope policy checks, audit events, red-team scenarios, and regression evaluation. Agree on release gates and a rollback plan before expanding autonomy.
- Safety & Compliance Layering
- Continuous Evaluation Systems
Evidence
Inspect the work before choosing a path.
- Inspect the architecture
Read the runtime layers alongside their implementation limits.
- Read the source
Follow repository links and setup guidance for a technical review.
- Compare product availability
Check each product's setup requirements and stated limitations.
- Open the documentation
Start with the guides, then follow the source behind the behavior.
The catalog describes local developer and document-workspace surfaces. Quasar is experimental and requires a separately running local service and provider configuration.
The film
See the direction. Inspect the evidence.
The films carry vision and roadmap context; they do not establish benchmarks or shipped capability.
The mark
The M is the stack: experience, then runtime, then memory.
Engagement process
Agree on the evidence before implementation.
01Audit
Proposed first step: inventory workflows, data, tools, and failure modes; agree on the risk register that guides the work.
02Design
Define retrieval, permissions, deployment topology, and an evaluation harness for the agreed scope.
03Build
Implement against the agreed harness on operator-owned hardware and establish a baseline the team can rerun.
04Harden
Plan red-team checks, rollback, observability, and operator training around the release gate.
Open development
Build in the open.
Explore public repositories, documentation, and plugins. Follow the source and contribute through each project’s own workflow.
Future vision
The next generation of human + AI.
These are product directions. Availability and implementation evidence remain in the source catalog.
MLAI Studio
A proposed workspace for visual agent design, prompt experiments, and inspecting context, memory, and execution.
Multimodal collaboration
A direction for working across text, images, audio, code, and spatial environments with explicit data boundaries.
MLAI Network
A proposed community where builders share agents, plugins, research, projects, and extensions.
Next step
Discuss your project.
Discuss a project to define its scope, data boundaries, and evidence requirements.