Open framework
Installable and local-first. Persistent memory uses project-local storage; optional dense retrieval gracefully falls back when its extras are absent.
TRW began with a practical problem: later AI-assisted sessions often had to rediscover requirements, decisions, and failure modes that a project had already encountered.
The framework preserves selected project context and delivery evidence so the next session has a better starting record—not a promise that every task automatically improves.
ORIGIN // FROM_CONTEXT_LOSS
The product grew by separating what belongs to a conversation from what a project should retain.
Useful constraints and decisions lived in conversation history. Later sessions often had to rediscover them.
Learning files and checkpoints made selected context available outside the conversation that produced it.
Requirements, reported repository checks, exception state, and handoff notes became separate inspectable records.
TRW packages those mechanisms behind MCP tools that can be used across supported coding clients.
MECHANISM // NOT_MAGIC
A stored learning can help a later session avoid rediscovery. It can also become stale. The useful mechanism is capture, recall, re-verification, and retirement—not an unbounded compounding claim.
Record a material finding with project context.
Keep it outside the conversation that produced it.
Surface eligible candidates for a later task.
Validate the candidate against the current repository.
Strengthen, supersede, or retire what changed.
NAME // THE_REAL_WORK
“The Real Work” emerged during early working sessions as shorthand for the part of engineering that generation alone does not solve: preserving the constraints, decisions, evidence, and hard-earned context that future work depends on.
The output is not only code. It is also the record that lets another person or agent understand what was attempted, checked, and left unresolved.
That is a product principle, not a claim of autonomous authorship or guaranteed compounding. TRW exists to make that record easier to create, inspect, and carry forward.
PRODUCT // WHAT_EXISTS
TRW connects project memory, workflow state, handoff contracts, and reported delivery evidence. It does not replace your coding client, repository checks, or engineering judgment.
Store material learnings and recall candidates in a later session.
BoundaryRecalled knowledge is context to re-check, not automatic truth.
Choose a lighter or deeper ceremony based on the task and preserve recovery state.
BoundaryPhase guidance is advisory; the coding delivery gate is the hard boundary.
Make ownership, artifacts, dependencies, and next actions explicit.
BoundaryYour coding harness executes helpers; TRW supplies the coordination contract.
Connect a requirement to implementation scope and caller-reported repository checks.
BoundaryTRW records check results; it does not independently execute or certify them.
EVIDENCE // CLAIM_BOUNDARY
Repository counts describe inventory and dogfooding activity. They do not prove that TRW improves outcomes on an independent workload.
Inspect the verification modelThe repository contains the storage, recall, workflow, receipt, and handoff paths described here.
TRW is used in its own repository. Inventory counts show activity and surface area, not efficacy.
Broad improvement across independent workloads remains an empirical question, not a marketing conclusion.
STATUS // AVAILABLE_SURFACES
TRW is built by Tyler Wall with extensive AI-assisted implementation and review. The product boundary matters more than the mythology: the open framework, hosted platform, and proprietary tester surfaces have different availability and data paths.
Installable and local-first. Persistent memory uses project-local storage; optional dense retrieval gracefully falls back when its extras are absent.
Opt-in organization-scoped services and dashboards when configured. This is separate from the open framework’s local operation.
Additional evaluation, orchestration, and research packages use a separate access path. They are not implied by the open install.
Install the local framework, capture a meaningful project constraint, and verify that a later session can retrieve and re-check it.