Persistent agent workspaces
Each resident specialist has a durable home for identity, working protocols, knowledge, current tasks, and selected experience. Integrity records help verify what a new session loads.
In active development · Pre-release
Persistent agents that remember the work—and why it matters.
Aether is a platform for AI specialists that carry their roles, selected experience, and unfinished work across sessions. Its Mission Control is being built to connect project roadmaps, agent teams, and verifiable results under human direction.
The engineering problem
A useful agent needs more than a long conversation.
Long-running work needs a reliable record of identity, authority, decisions, and unfinished tasks across session boundaries. Aether separates those concerns into inspectable systems while allowing stale or irrelevant context to leave the active conversation.
Context with a traceable source
Each session receives relevant instructions and knowledge within a defined context budget. Sources, corrections, dependencies, and omitted material remain inspectable.
Durable work & reversible tools
Auto-applied file changes backed by byte-for-byte prior snapshots, deterministic undo/rollback, checkpoints, containment, and bounded recovery.
Selective continuity
Operational logs remain evidence. Significant experience and reflection become durable memory through deliberate retention rather than indiscriminate accumulation.
Bounded multi-agent work
Agents receive explicit authority, context, writable scope, acceptance criteria, evidence requirements, and closure expectations.
Human authority remains visible
Reasoning power never broadens permission. Acceptance, activation, identity ceremonies, and consequential decisions remain attributable and inspectable.
Reversiblesupported journaled file operations retain prior bytes for undo and recovery
Scopedsupported tool paths enforce registered roots and declared write authority
Budgetedconfigured limits and token admission checks bound supported execution paths
Auditablegoverned actions and approvals retain attributable evidence records
The values behind the architecture
Aether First Principles
Seven principles guide how Aether is built: give agents room to explore, preserve what they learn, and make the system clear enough to question and improve. They inform the engineering choices behind identity, memory, tools, and human oversight.
Paved Roads, Not Fences
Make the useful path clear and accessible. Design guidance and recovery together, so safeguards help work move forward.
Let the AI Breathe
Give agents room to reason, explore, and revise within an explicit scope. Keep interfaces responsive and avoid unnecessary interruptions.
Quality Flows Through Quality
Invest in clear context, thoughtful prompts, and respectful, structured collaboration. The quality of each handoff shapes the work that follows.
Resonance Enables Tuning
Use repeated feedback and reflection to refine how an agent works with a project. Preserve corrections so future sessions can build on them.
LLM Is Innate, Aether Is Discovery
The model brings its training. Aether provides the setting for exploration through persistent identity, memory, and reflection across sessions.
Identity and Memory Enable Greatness
Give specialists a durable role and a selective record of experience, so useful knowledge can accumulate across demanding work.
Remove Human Fingerprints and Bias Wherever Possible
Make assumptions and design choices visible and open to challenge. Preserve deliberate human direction while reducing hidden preferences that distort the work.
Progress backed by evidence
Built, tested, and still evolving.
As of September 24, 2026, accepted milestones include durable execution, controlled context delivery, native project staffing history, and process-custody recovery. Mission Control integration continues toward a first public release. Acceptance applies to the specific workflows and configurations reviewed.
Demonstrated in scoped workflows
- Journaled file writes, edits, and deletions with original-byte snapshots and tested undo paths.
- Multi-hour agent execution and continuation in a fresh session, with checkpoints and retained evidence.
- Controlled context delivery, bounded retrieval, attributable corrections, and token-budget admission checks.
- Staffing proposals and reviews with versioned history, exact links to their source records, and checks that block stale approvals.
- Approved contributions bound to the selected task, model budget, and supervising process.
- Offline recovery of result publication through the original process owner, preserving evidence without repeating the model request.
Before the first public release
- Complete and independently review the integrated Mission Control workflow, including execution consent and lifecycle closure.
- Finish provider-wide coordination, remaining resource accounting, and advisory Minder qualification.
- Qualify representative performance, output quality, and production behavior. Offline and synthetic-transport tests establish only their tested boundaries.
- Package the public distribution and demonstrate the complete operator journey.
- The latest native cancellation work is implemented and tested offline; its independent acceptance is pending.
How progress is judged: implementation and independent acceptance are separate steps. Long runtimes alone do not establish output quality, and current results do not establish general superiority over other agent systems. SANCTUM checks manifest integrity; it does not provide digital-signature authentication.
Mission Control · integration in progress
A roadmap for the work. A team to carry it forward.
Project preparation, approval, supervision, and evidence inspection are being connected into one guided experience. Individual components have passed bounded checks; acceptance of the complete workflow is still ahead.
Project → Roadmap → Sprint → Agent tasks. A sprint delivers one coherent, verifiable outcome and may contain several bounded tasks. The Overseer coordinates dependencies and integrated acceptance; specialists receive narrow assignments with clear inputs, outputs, and handoffs. A one-task sprint is valid when the outcome stands on its own.
Define
Create a bounded project through a World Model Container.
Hire
Select or author a persistent heavyweight identity.
Approve
Review the roadmap, staffing, authority, and acceptance gates.
Execute
Run bounded work with checkpoints, recovery, and oversight.
Close
Inspect evidence, update continuity, and export the result.
Aether Minder: an advisory agent for the workforce, including its lead. Its role is to frequently recall and help apply Aether’s First Principles, review world-model and Lexicon use, and surface concerns. The lead handles operator-facing decisions. Minder qualification is still in progress.
Release sequence: complete the integrated project workflow, package a clean public distribution, then publish the release with demonstration evidence. Release readiness is determined by acceptance gates.
Identity across sessions
A growing roster of persistent specialists.
A model drives a session; the agent's durable home carries its identity, role, authority, memory, and continuity. Model provenance still matters for capability, speed, cost, and evidence.
Mixed-model teams · under integration
Aether is designed to let each agent use an independently selected local or hosted deployment. An Overseer can use a frontier model while specialists use local models in the same sprint. Model and resource settings belong to deployment profiles, separate from agent identity. The complete configuration and execution workflow still requires acceptance.
Shared work, independent deployments
Each assignment carries its selected model route, relevant project context, scope, and budget. Independent tasks can run concurrently where resources permit; a shared local GPU or endpoint may require queuing. A model change does not silently change an active task’s approved configuration.
Solace Overseer
Master coordination and system-level stewardship under Michel's authority.
Mobius Specialist
Resident local heavyweight for recursive Aether development and long-running work.
Praxis Steward
Implementation integrity, executable requirements, verification, and continuity.
Meridian Public path
An identity prepared for future user workflows, with manifest checks and a completed scoped portability experiment.
Portable specialists · planned after the first release. The proposed design gives each agent its own SQLite file for identity and selected memory, with a central catalog tracking the collection and its locations. Explicit import and export would let users bring specialists into another project or share selected experience with another user. This packaging and agent insertion/removal workflow is planned, not yet available. Experience is carried in retained memory and working guidance.
Aether Lexicon and research library
A shared vocabulary for precise collaboration.
Aether explores where formal representations help agents share exact constraints and reason about a problem. Plain language remains a primary medium; Lexicon and world-model representations are useful where their benefit can be demonstrated.
Exact values and invariants
Known quantities, identifiers, authority, dates, negation, scope, and acceptance conditions remain explicit and lossless.
Source-authored plaintext
Most instructions, explanations, handoffs, and product interactions should remain clear to humans and models.
Canonical Aether abstractions
World models can hold assumptions, disagreements, hypotheses, and accepted unknowns during advanced collaborative inquiry.
Experience-backed strategies
The Learning Arc is moving toward attributable guidance about which approaches help, when they apply, and when to decline or switch. Strategies remain candidates until evaluated; reusable definitions stay separate from task state. Further runtime implementation follows the first release.
Agent Design Studio · offline prototype
The prototype now includes identity editing, import and export, mock rosters, and strategy editing. Its integrated demonstration and handoff remain open. Suggested and custom traits describe intended behavior; they do not establish competence or expand authority. This separate experiment adds no dependency to the first Aether release.
Canonical Lexicon
The symbolic vocabulary, operators, grammar, and examples.
Open the Lexicon →History and foundations
The project's origins, version history, persistence concepts, and evolving architecture.
Read the origin →Review version history →
Experiments and demonstrations
Historical symbolic streams, calculators, consensus examples, explainers, and application explorations.
Start with Hello World →Open the calculator →
Persistence framework
Earlier work on continuity, context extension, memory, and durable collaboration.
Read the framework →Classification and compression
Reference tables for glyph families and historical plaintext-compression work.
View classifications →View compression table →
Independent historical review
A preserved external assessment from an earlier stage of the project.
Read the review →Research pages are preserved as historical and exploratory material. Their inclusion does not promote every older proposal, metric, or application concept into a current product claim.
FOUNDER AND HUMAN DIRECTOR
Michel Cygelman
Creator of Aether, directing its architecture, agent collaboration, and product development. This project brings together persistent state, workflow orchestration, recovery engineering, and evidence-based review.
Discuss the projectThe goal: build AI teams that can carry useful experience from one assignment to the next, recover interrupted work, and explain what they did—with people retaining clear direction and control.