haylan/LLM-Server
1
0 Fork 0
Code Issues 4 Pull Requests Actions Packages Projects Releases Wiki Activity

Target doc structure #60

New Issue
Closed
opened 2026-09-24 19:08:21 +00:00 by haylan · 1 comment
haylan commented 2026-09-24 19:08:21 +00:00
Owner
Copy link
Copy source

Part of #55

Question

Which doc files exist after cleanup, what each explains, and which source each one links as its source of truth? Includes the fate of stale docs (e.g. docs/proxy-request-priority.md) and the README's role. HITL: prototype an outline to react to.

Part of #55 ## Question Which doc files exist after cleanup, what each explains, and which source each one links as its source of truth? Includes the fate of stale docs (e.g. docs/proxy-request-priority.md) and the README's role. HITL: prototype an outline to react to.
haylan added the wayfinder:prototype label 2026-09-24 19:08:21 +00:00
haylan added a new dependency 2026-09-24 19:08:33 +00:00
#57 Research-doc index: current vs. superseded
haylan added a new dependency 2026-09-24 19:08:33 +00:00
#58 Removal inventory: image generation and Open WebUI
haylan added a new dependency 2026-09-24 19:08:33 +00:00
#59 Glossary: pin down domain terms in CONTEXT.md
haylan added a new dependency 2026-09-24 19:08:33 +00:00
#61 Comment destination map
haylan added a new dependency 2026-09-24 19:08:34 +00:00
#62 Agent guide: writing docs and extending the glossary
haylan added a new dependency 2026-09-24 19:12:23 +00:00
#63 Intent vs. source: fix the config or fix the doc?
haylan self-assigned this 2026-09-24 19:29:09 +00:00
haylan commented 2026-09-24 19:31:27 +00:00
Author
Owner
Copy link
Copy source

Resolution — target doc structure

Every doc opens with its Source of truth line (map Notes); terms per CONTEXT.md.

Path Purpose Source of truth Absorbs
README.md One-paragraph description, Quickstart (./scripts/update.sh), index of every doc docker-compose.yml, scripts/update.sh — (gateway/LiteLLM/stealth asides move out)
CONTEXT.md Glossary — done in the Glossary ticket
docs/architecture.md One file, a section per service (llama-server, qwen-classifier, omniroute, qdrant): why it exists, how backend → provider connection → gateway → workload fits docker-compose.yml → each service GPU_MAX_HW_QUEUES/ROCm bug; numeric GIDs (4 copies → 1); classifier rationale + VRAM story (resolved in favour of the compose account); --cache-reuse; per-slot context math (3 copies → 1); n-gpu-layers/mmap; healthcheck; mem_limit
docs/gateway.md Replaces proxy-key-onboarding.md. Provider connections live in the dashboard; API key create/hand-over/revoke; the three OmniRoute timeouts (one section → the 3 research docs); SearXNG web search + search.home gotcha; OmniRoute stealth/MITM caution docker-compose.yml → omniroute, .env.example key-mint/#37 story, timeout duplicates, provider-config-in-dashboard duplicates
docs/network-access.md Reachability per port: gateway API (LAN + reverse-proxy hostnames), dashboard (LAN, not internet-routed), backends (no host port), qdrant (loopback) docker-compose.yml → ports/expose "llama-server internal-only" (6 copies → 1)
docs/operations.md What update.sh does and why: .env sync/conflict prompt, secret generation (storage-key warning stays inline in source too), host-resolved values, vendored gum, model downloads; manual deploy steps scripts/update.sh, .env.example update.sh header/body comments
docs/coding-cli-setup/ index + 4 per-CLI pages kept; ports/terms fixed; tool-calling risk stated once (index); classifier rationale → link to architecture.md each CLI's own config --jinja duplicate, tool-calling risk (3 → 1)
docs/server-planing.md Kept unchanged — —
docs/research/README.md New index: current / superseded-by / drift note per doc (from the Research-doc index ticket) — —
docs/agents/ issue-tracker.md, domain.md, + writing-docs guide (Agent guide ticket) — —

Moves / deletes

  • docs/proxy-request-priority.md → docs/research/, marked superseded by the OmniRoute migration.
  • docs/proxy-key-onboarding.md → becomes docs/gateway.md.
  • Delete the two image-generation research docs.

Build plans: neither server-planing.md nor research/500k-context-2-parallel-agents-gpu-roadmap.md is an active plan — both are labelled background research in the indexes; files unchanged.

## Resolution — target doc structure Every doc opens with its *Source of truth* line (map Notes); terms per `CONTEXT.md`. | Path | Purpose | Source of truth | Absorbs | |---|---|---|---| | `README.md` | One-paragraph description, Quickstart (`./scripts/update.sh`), index of every doc | `docker-compose.yml`, `scripts/update.sh` | — (gateway/LiteLLM/stealth asides move out) | | `CONTEXT.md` | Glossary | — | done in the Glossary ticket | | `docs/architecture.md` | **One file**, a section per service (llama-server, qwen-classifier, omniroute, qdrant): why it exists, how backend → provider connection → gateway → workload fits | `docker-compose.yml` → each service | GPU_MAX_HW_QUEUES/ROCm bug; numeric GIDs (4 copies → 1); classifier rationale + VRAM story (resolved in favour of the compose account); `--cache-reuse`; per-slot context math (3 copies → 1); n-gpu-layers/mmap; healthcheck; mem_limit | | `docs/gateway.md` | Replaces `proxy-key-onboarding.md`. Provider connections live in the dashboard; API key create/hand-over/revoke; the three OmniRoute timeouts (one section → the 3 research docs); SearXNG web search + search.home gotcha; OmniRoute stealth/MITM caution | `docker-compose.yml` → omniroute, `.env.example` | key-mint/#37 story, timeout duplicates, provider-config-in-dashboard duplicates | | `docs/network-access.md` | Reachability per port: gateway API (LAN + reverse-proxy hostnames), dashboard (LAN, not internet-routed), backends (no host port), qdrant (loopback) | `docker-compose.yml` → ports/expose | "llama-server internal-only" (6 copies → 1) | | `docs/operations.md` | What `update.sh` does and why: .env sync/conflict prompt, secret generation (storage-key warning stays inline in source too), host-resolved values, vendored gum, model downloads; manual deploy steps | `scripts/update.sh`, `.env.example` | update.sh header/body comments | | `docs/coding-cli-setup/` | index + 4 per-CLI pages kept; ports/terms fixed; tool-calling risk stated once (index); classifier rationale → link to architecture.md | each CLI's own config | `--jinja` duplicate, tool-calling risk (3 → 1) | | `docs/server-planing.md` | Kept unchanged | — | — | | `docs/research/README.md` | **New** index: current / superseded-by / drift note per doc (from the Research-doc index ticket) | — | — | | `docs/agents/` | issue-tracker.md, domain.md, + writing-docs guide (Agent guide ticket) | — | — | **Moves / deletes** - `docs/proxy-request-priority.md` → `docs/research/`, marked *superseded by the OmniRoute migration*. - `docs/proxy-key-onboarding.md` → becomes `docs/gateway.md`. - Delete the two image-generation research docs. **Build plans:** neither `server-planing.md` nor `research/500k-context-2-parallel-agents-gpu-roadmap.md` is an active plan — both are labelled *background research* in the indexes; files unchanged.
Sign in to join this conversation.
No Branch/Tag Specified
Labels
Clear labels
wayfinder:grilling
Wayfinder grilling ticket
 wayfinder:map
Wayfinder map issue
 wayfinder:prototype
Wayfinder prototype ticket
 wayfinder:research
Wayfinder research ticket
 wayfinder:task
Wayfinder task ticket
No labels wayfinder:prototype
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
haylan (Arthur Erlich)
No Assignees haylan
1 Participants
Notifications
Due Date
No due date set.
Blocks
#61 Comment destination map
haylan/LLM-Server
#62 Agent guide: writing docs and extending the glossary
haylan/LLM-Server
Depends on
#56 Audit: contradictions and duplicates across docs and code comments
haylan/LLM-Server
#57 Research-doc index: current vs. superseded
haylan/LLM-Server
#58 Removal inventory: image generation and Open WebUI
haylan/LLM-Server
#59 Glossary: pin down domain terms in CONTEXT.md
haylan/LLM-Server
#63 Intent vs. source: fix the config or fix the doc?
haylan/LLM-Server
Reference: haylan/LLM-Server#60