Research mode switches to nine kind-generic tools: search_files,
get_file, read_content, search_zims, get_zim_entry, get_inventory,
list_files, read_file, search_places. Tool definitions drop from 20
tools / 8,947 chars (~2,240 tokens) to 9 / 3,881 (~970) on every
request, a sixth of the 8k context the small tier runs with instead of
a quarter. The remaining descriptions are trimmed toward 300 chars.
The replaced per-kind endpoints are removed (four searches, three
getters, three readers, search_suggestions, the slug inventory route);
their tests move onto the generic routes. The research prompt, its
worked example, and the empty-search hint no longer tell the model to
list collections first: channels resolve by name from the library map
that is already in the prompt. list_collections, list_tags, and
get_map_overview stay as endpoints for the MCP wrappers but leave the
chat allowlist; the MCP search_suggestions tool goes since search_files
returns the same matches inline.
scripts/ai_eval.py expects search_files and treats "what channels do I
have" as answerable from context. On the 1.7B model against the dev
library both catalogs score 14/18, but the compact one answers with one
tool call where the old one needed up to five, no longer invents the
example's channel, and keeps kinds straight when asked for videos.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Small models carry every tool definition on every request, and the
per-kind trios (four searches, three getters, three readers) cost a
quarter of an 8k context while giving the model near-duplicate choices.
Three generic endpoints replace them:
- POST files/search: one search over videos, archives, and docs with an
optional kind; channel (name or id), domain, author, subject each
imply their kind and are ignored for other kinds. When the term also
names channels/domains/authors, they are returned as `matches` so the
model can narrow without a second tool. Channels resolve by name, so
the library map already in the prompt is enough: no list call first.
- GET files/{id}: detail by id whatever the kind; videos now report
has_captions/has_comments and the full description, archives keep
their snapshot history.
- GET files/{id}/content?part=text|comments: captions, page text, or
comments, paged; a wrong part for the kind says what the file is.
Inventories take an optional slug on GET /inventories (one tool lists
and fetches). Listings are slimmed: no derivable captions/text links,
no size/mimetype, duration as H:MM:SS, published as a date. The
library map gains the tag names.
wrolpi_mcp keeps every existing tool name; the wrappers now call the
generic routes, and search_files/get_file/read_content are exposed too.
The old per-kind endpoints remain until the modes switch over.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Five new read-only /api/ai endpoints, each with a matching wrolpi_mcp
proxy tool: list_tags (tags with what they are applied to, plus recent
names), search_suggestions (matching channels/domains/authors/subjects
and file/Zim result estimates), list_downloads (queue summary and lean
recurring/one-time entries with relative destinations; status filter;
errors keep their tail, which holds the exception message),
get_map_overview (map files, search-index state, subscriptions, pins
with map links), and search_places (place search with map links and
optional proximity ranking).
Research mode allowlists the library-facing four. Nothing here can
start, stop, or retry a download.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
New read-only GET /api/ai/files/list lists one directory of the media
directory (directories first, then files with size and guessed
mimetype), paged at 100 entries. Every entry carries its relative
path for list_files or read_file. Same guards as read_file via a
shared resolver: no escaping the media directory, and the config
directory is refused and hidden from the root listing.
The agent loop picks the operation up from the spec; library mode
allowlists it. wrolpi_mcp gains the matching list_files proxy tool.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The re-run showed "Is everything running?" parroting the example's
all-clear without checking. The system prompt now states that state is
never known from memory or examples — every health answer starts with a
fresh get_system_status — and the example's answer cites its own
example data (service names, version) so a verbatim parrot is
self-evidently wrong.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The eval's two remaining System misses were the model investigating a
fictional failed wrolpi-kiwix — planted by the few-shot example itself.
The example now shows a healthy status and an all-clear answer; it
still demonstrates status-first without seeding a fake diagnosis.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The first scored eval run (39/45) caught two example defects: a user
asking about the help example's own topic got the example's invented
UI detail parroted back without a search, so the example now carries
true content from the real certificates doc on a topic users rarely
lead with; and list-style answers dropped WROLPi links because the
research example only showed links in prose, so its answer now models
a numbered list with a link per item.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Response-quality work, all aimed at giving the model enough context to
answer accurately:
- Every chat's system prompt now carries a live library map (today's
date, counts, and the exact names of channels, domains, zims, and
inventories, capped and cached 60s). Exact spellings are in context
before the first token, and the list_collections round-trip — minutes
on slow hardware — is usually unnecessary.
- The request is budgeted against the model's real context: oldest
turns are trimmed first, the current question is never dropped, and
an over-budget turn gets a clear "start a new conversation" error
instead of a silently truncated prompt. cache_prompt reuses the KV
cache for the stable prefix, which is most of the latency on a Pi.
- context_size plumbing fixed: selecting a model adopts its catalog
default (the 4B was running at 8k instead of 16k), and llama-server
runs --parallel 1 so the single user gets the whole context.
- temperature is configurable in ai.yaml (default 0.3 — small models
call tools far more reliably near 0.2-0.3 than at server defaults).
- One worked tool-use example per mode is prepended to every
conversation; small models imitate far better than they obey.
- scripts/ai_eval.py scores the assistant against a live WROLPi on
mechanical checks (finished, called the right tool first, included a
WROLPi link, no external URLs) so prompt changes stop being validated
by anecdote.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Live testing surfaced three ways small models went wrong summarizing a
library: they gave up after one empty search (misspelled channel names
found nothing), so zero-result searches now carry a retry hint and the
Research prompt requires list_collections before channel/site questions
and search_videos(channel_id) before claiming a channel is summarized;
they presented the familiar YouTube source URL instead of the WROLPi
link, so listings omit the source URL (archive detail keeps it); and
they invented links for channels, so channel collections now carry
their real /videos/channel page link. Search descriptions state that
search_str is optional for browsing the newest items.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three findings from running Qwen3-1.7B against the docker stack:
the llm image lost libgomp when the build tools were removed
(llama-server needs the OpenMP runtime); the Help prompt let the
small model answer from memory instead of the docs, so it now
requires search_help before any answer; and help search results
called the slug "doc" while get_help_doc's argument is "slug", so
the model invented slugs — the names now match.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Tool definitions are generated in-process from the /api/ai blueprint's
own OpenAPI operations (operationId = tool name, description = tool
description, dataclass bodies + declared parameters = one flat JSON
schema) — the spec IS the catalog, nothing to drift. Manage endpoints
never become tools. A mode is a server-enforced system prompt + tool
allowlist (Help / Research / System per ai.md); the executor can only
reach allowlisted /api/ai endpoints, path parameters are encoded and
traversal-rejected so a crafted argument cannot escape an endpoint.
POST /api/chat streams SSE (status/token/tool_call/tool_result/done/
error), starts llama-server on demand, validates tool args server-side
with instructive errors the model can retry on, caps tool results with
a paging hint, and bounds the loop by ai.yaml max_tool_calls plus a
hard timeout. Caddy passes the stream unbuffered (flush_interval -1)
in both deployments.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
