Continue after compaction
Open a new session without reconstructing the architecture, repeating corrections, or rediscovering unfinished work.
Coding Accelerator is the persistent intelligence and coordination layer underneath AI-assisted development. It preserves decisions, constraints, rejected approaches, investigations, unfinished work, and current project state outside any assistant's temporary context—then prepares relevant Project Truth for the next session and helps multiple agents work without losing alignment.
Compaction forgets. Coding Accelerator remembers.Open a new session without reconstructing the architecture, repeating corrections, or rediscovering unfinished work.
Keep the reasoning, evidence, status, and affected areas attached to important project decisions.
Make rejected and superseded approaches visible before an assistant confidently implements them again.
Inspect agents, sessions, tasks, branches, changed files, possible overlap, stale assumptions, and handoff state.
Review the project intelligence selected for hydration instead of relying on an unsupported generated summary.
Distinguish current state from what is historical, rejected, superseded, unresolved, or conflicting.
The assistant knew why authentication belonged at the gateway. It knew which migration remained unfinished, which production fix failed, and why the most obvious alternative had already been rejected. Then the session compacted.
The files remained, but the working reasoning disappeared.
You became the restore process.
Coding Accelerator gives that knowledge a durable home outside the temporary session. Your assistant can start over. Your project intelligence does not.
Keep authentication in API gateway.
Reason:Internal services must never validate external identity tokens directly.
Rejected:Per-service authentication.
Claude Code can refactor a schema while Codex updates the API handler and Copilot builds the interface against yesterday's contract. Each assistant may be individually competent while the combined work still conflicts.
Switchboard brings agents, sessions, tasks, branches, changed files, decisions, and possible conflicts into one inspectable project operation.
Copilot is using PaymentRequest v2. Claude and Codex moved the shared contract to v3.
Coding Accelerator captures the knowledge surrounding the code—not merely another transcript. Project Truth makes current, rejected, superseded, and conflicting information explicit.
Evidence-backed retrieval finds what matters for the task. Intelligent hydration supplies a focused working set to the next session without dumping the entire project history into its context window.
This intelligence remains available across compaction, restarts, terminals and agent providers.
Switchboard coordinates the MANY. Agents, sessions, tasks, branches, changed files, shared decisions, and possible overlap become one inspectable project operation.
Coding Accelerator is not another coding agent. It is built to give your agents a shared project memory and coordination layer.
Not five disconnected features. One intelligence cycle that gives the project durable intelligence of its own.
Evidence-backed retrieval finds the project facts that matter for the task being opened.
Coding Accelerator is designed to sit underneath the assistants you already use. It preserves, retrieves, and coordinates project intelligence; it does not generate code in their place.
Maintain an inspectable record of the project's current decisions, constraints, requirements, risks, and unresolved questions.
Retain the reasoning surrounding the code across compaction, restarts, terminals, and new sessions.
Record what was decided, why it was decided, what was rejected, and when a later decision superseded the original.
Retrieve relevant project intelligence with its source, status, reasoning, and affected areas attached.
Prepare task-relevant context for an incoming session while keeping irrelevant project history out of the working set.
See which assistants are active, what they are doing, where they are working, and which decisions affect that work.
Inspect possible changed-file overlap, stale assumptions, divergent contracts, duplicated investigations, and conflicting workstreams.
Review capture freshness, search coverage, unresolved conflicts, synchronization state, and hydration readiness through measurable conditions.
Understand which tools, repositories, accounts, and storage surfaces participate in the project-intelligence flow.
The $299 one-time purchase provides a perpetual native Windows desktop license for Coding Accelerator.
$299 once. No recurring subscription is required for the desktop license.Protocol Loom is presented separately as post-MVP platform direction. Its future SYSTEM-level capabilities are not represented as part of the current desktop product.Persistent project intelligence
✓Project Truth
✓Context continuity
✓Decisions and rejected-approach memory
✓Evidence-backed retrieval
✓Intelligent session hydration
✓Agent, session, and task awareness
✓Switchboard multi-agent coordination
✓Conflict and overlap visibility
✓Context Health
✓Integration and storage visibility
✓Native Windows project workflow
✓The codebase isn't the whole system.
A single feature may cross frontend applications, APIs and routes, backend services, schemas, databases, queues, event systems, WebSockets, webhooks, infrastructure, cloud resources, identity, SaaS, B2B, external partners, and deployment relationships.
No individual coding session sees that entire path. The connections—and the technical rules those connections depend on—matter too.
Protocol Loom is the post-MVP architecture beyond Coding Accelerator's core project-memory and multi-agent coordination layer.
It is designed as a non-owning observation, topology, tracing, contract, verification, and continuity layer around the systems you already own.
One end-to-end functional execution or connectivity path through the system.
Coding Accelerator is the commercial product.
Switchboard is the project-intelligence and multi-agent coordination engine operating inside the project workflow.
Protocol Loom is the non-owning observation and verification wrapper around the broader technical project/system surface.
Protocol Loom is intended to connect a changed route, schema, token expectation, certificate reference, or event envelope to the Threads and Splices that depend on it—using permitted evidence, not an invented architecture graph.
Future-facing, evidence-backed diagnosis—without invented topology or automatic ownership of your systems.
Gateway configuration · Token metadata · Payments API policy
These are governing design boundaries for the future Protocol Loom architecture, not claims about current system-observation availability.
Observe permitted evidence without becoming the authority that controls application business logic or the systems being examined.
A connection appears only when available evidence supports it—not because an AI guessed two boxes probably communicate.
Map safe references, expectations, and metadata without ingesting actual secret, key, credential, private-key, or bearer-token values.
Know that a token exists.
Never need to know the token.
Walk through five core-product surfaces, then inspect the separately labeled Protocol Loom concept preview.
Current objective, active agents, measurable Context Health, and attention items in one view.
Knows what happened in one conversation.
Knows what exists in files.
Knows what someone explicitly documented.
Maintains structured intelligence about why the project is in its current state and makes it available across AI-assisted work.
Source control preserves code history.
Coding Accelerator preserves project reasoning.
Move from a compacted session to a new one without reconstructing decisions, constraints and unfinished work.
Coding Accelerator is the continuity and coordination layer underneath your AI-assisted development stack—not a replacement for it.
INTEGRATION TARGETDesigned for interoperability with Claude Code while keeping Claude Code as the coding agent.
INTEGRATION TARGETDesigned for interoperability with Codex so relevant decisions, constraints, investigations, and active work can follow the task.
INTEGRATION TARGETDesigned for interoperability with Copilot-assisted work against the same current Project Truth used across the project.
INTEGRATION TARGETDesigned for interoperability across the editors and development environments already in use.
INTEGRATION TARGETDesigned to connect tasks, branches, changed files, and possible overlap to repository work.
INTEGRATION TARGETDesigned to make participating project-intelligence and connected-service boundaries visible while accounts remain under your control.
Keep project intelligence visible on the Windows desktop and connected to the project surfaces you control.
Bring permitted storage, API, and cloud accounts without turning Coding Accelerator into the owner of those systems.
Optional managed services may be offered independently. They are not required by the $299 perpetual desktop license.
The project itself should remember. AI coding assistants can be remarkably capable inside the session in front of them. But software development lasts longer than a session.
A developer establishes architecture, constraints, corrections, debugging discoveries, rejected approaches, unfinished work, and the reasoning behind important decisions. After compaction, restart, terminal closure, or a new session, the developer becomes the restore process—explaining the same project again.
More agents create a second continuity problem. Claude Code, Codex, Copilot, IDE agents, and other assistants can each be capable while working from different project context, producing duplicate investigations, stale assumptions, conflicting changes, revived decisions, and poor handoffs.
Coding Accelerator exists because this recurring friction should be engineered out of AI-assisted development. The project should have durable intelligence of its own. The same principle eventually extends beyond sessions and agents to the technical system itself.
Your project mind should not disappear when you change models, tools, terminals, or IDEs.
Important project intelligence should carry its status, source, reasoning, and affected work—not merely a confident summary.
Your storage, connected accounts, security boundaries, and technical systems remain yours.
Direction is not availability. No future capability on this page should be read as a release date, entitlement, or current product claim.
Persistent project intelligence, Project Truth, memory, evidence-backed retrieval, intelligent hydration, agent/session/task awareness, Context Health, and multi-agent coordination.
ONE + MANYEvidence-backed observation, topology, tracing, contract verification, change impact, and continuity across permitted source, configuration, infrastructure, runtime, cloud, SaaS, B2B, identity, messaging, and external integration surfaces.
SYSTEMWhat you can verify now—through visible project state, evidence, decision history, hydration context, and coordinated workstreams.
Start a new session and inspect the project intelligence selected for restoration.
NEW SESSION → RESTORED CONTEXTOpen a current decision and inspect its evidence, reasoning, status, and affected work.
DECISION → EVIDENCE → IMPACTReview a rejected or superseded approach instead of letting an agent revive it as new.
REJECTED · SUPERSEDED · CURRENTOpen multiple workstreams and inspect possible overlap, stale assumptions, and handoff state.
AGENTS → WORKSTREAMS → CONFLICTSReview the context selected for an incoming task before it reaches the next agent session.
TASK → RETRIEVE → REVIEW → DELIVERInspect current project truth directly instead of relying on an unsupported generated summary.
STATUS · SOURCE · REASONING · SCOPEContext loss costs more than the minutes spent retyping instructions. It also creates repeated investigations, reopened architecture debates, revived approaches, conflicting changes, stale review cycles, and manual coordination across parallel agents.
At an effective engineering cost of $100 per hour, that is approximately three hours. At $150 per hour, it is approximately two.
Illustrative arithmetic—not a promised productivity result.Buy the tool. Keep the tool.
Stop paying the context-rebuild tax.
$299 once for a perpetual native Windows desktop license. No recurring subscription is required for the desktop software. Licensed for use on one active device at a time. Transferable to a replacement device.
Perpetual Coding Accelerator desktop license.
What Coding Accelerator is, what it replaces, and where the platform goes next.
No. It does not replace Claude Code, Codex, Copilot, or another coding agent. It is the persistent project-intelligence and coordination layer beneath AI-assisted development.
Instruction files are useful, but they are usually manually maintained snapshots. Coding Accelerator structures decisions, evidence, status, rejected approaches, investigations, current work, and task relevance.
The repository shows what exists in files. It rarely explains why an approach was selected, what failed previously, which alternative was rejected, or what remains unfinished.
No. Git preserves code history. Issue trackers preserve documented work. Coding Accelerator preserves and coordinates the project reasoning surrounding AI-assisted development.
Yes. ONE is the first problem. Project knowledge should survive compaction, restarts, and new sessions even when only one assistant is involved.
MANY introduces duplication, overlap, stale assumptions, contradictory decisions, and poor handoffs. Switchboard makes those workstreams inspectable against shared Project Truth.
The core product centers on persistent project intelligence, Project Truth, context continuity, evidence-backed retrieval, intelligent hydration, agents, sessions, tasks, Switchboard coordination, conflict awareness, Context Health, and integration and storage visibility.
No. It is a one-time purchase for a perpetual native Windows desktop license.
Coding Accelerator makes local and connected storage boundaries visible so you can understand where project intelligence lives and which accounts participate. You may bring your own AI, API, storage, and cloud accounts; optional managed services, when offered, are separate.
No. Protocol Loom is post-MVP, exploratory platform direction. It describes how Coding Accelerator may expand from project intelligence into broader SYSTEM-level observation and verification.
Its governing design boundary is to map safe references, expectations, metadata, and contract information—never actual secret, key, credential, private-key, or bearer-token values.
Use the guided product model below as the documentation entry point for installation, Project Truth, hydration, Switchboard, storage, and security boundaries.
Register a project, confirm storage visibility, capture the first decision, and hydrate a new agent session from current Project Truth.
INSTALL → REGISTER → CAPTURE → HYDRATEStop reconstructing context and coordinating coding agents by hand. Give the project durable intelligence that can outlive any one assistant session.
Many agents. One project mind.Windows desktop · One-time purchase · Perpetual license