Retrieval Gate
Enter through a map, search, source paths, or a known topic. Stop as soon as the task has enough.
tree → context → showPersistent external context · V3
Give your codebase a memory that lasts.
llmdoc preserves architecture, constraints, and working knowledge that source code cannot cheaply give back—so coding agents reach the Context Floor faster.
Time to Context Floor
Reach enough context to solve the problem.
The Context Floor is the richness of context an agent needs for the task at hand. llmdoc does not pour a repository into the window; it reveals a map, then the relevant knowledge.
01ls → find → rgmap the repository
02read × manyinfer boundaries
03search againrecover missed relations
04context?still reconstructing
Tool callsRepeated · broad
Context footprintKeeps growing
Signal densityVariable
01llmdoc treeread the knowledge map
02llmdoc context --fileslocate knowledge owners
03llmdoc showload only what matters
04context ✓begin the work
Tool callsIntent-routed
Context footprintBounded
Signal densityFocused
Knowledge pipeline
Three gates decide what enters context.
Reflection Gate
Capture high-signal corrections and verified failures—not transcripts, trivia, or every turn.
signal → candidateStable Knowledge Gate
Promote only durable conclusions that change future choices and are expensive to reconstruct.
verify → llmdoc/The boundary
Knowledge, not inventory.
Keep
- Architecture decisions and rationale
- Boundaries and cross-module contracts
- Invariants and non-obvious failure semantics
- Risky, repeatable workflows
Leave in canonical sources
- Source and API inventories
- Logs, temporary plans, session summaries
- Facts cheaply rebuilt from schema, tests, or help
- Unverified speculation
Start at the map
No dependency added to your project.
Run the CLI as external tooling. npx uses the machine cache without changing the consumer repository.
npx -y @tokenroll/llmdoc tree
npx -y @tokenroll/llmdoc context --files src/api/client.tsMarkdown / MDXGit-native validityClaude + Codex
Installation and host setup →Understand / retrieve / verify
Understand the model. Retrieve by intent.
The documentation explains llmdoc's concepts, workflows, and safety boundaries. The CLI handles deterministic retrieval, status checks, and knowledge maintenance.