## Background
This branch started as a focused fix to agentic RAG regexp retrieval
semantics (`f80556585`) and grew into the full agentic RAG path. The
title no longer describes the contents, so it has been rewritten.
The PR now covers three largely independent lines of work:
### 1. The agentic RAG is reachable from the UI
`internal/agentic_rag` (the eino-ADK ReAct explorer) was already built
and wired, but only reachable by hand-crafting an `agent_mode` kwarg. It
is now the sixth option in the chat mode selector (`reasoning` level 5).
One subtlety worth stating plainly: **levels 1-4 and level 5 are not the
same agent.** Levels 1-4 go through `internal/rag/agentic-rag` (the
harness graph) with a depth chosen by `harnessModeForLevel`; level 5
switches engines outright to `internal/agentic_rag`. That is why level 5
must never reach `harnessModeForLevel` — its `level >= 4` case would
silently answer "ultra" for a level outside its domain.
### 2. Per-dialog failover chain
`agenticModelChain` resolved exactly one model and the caller then used
`chain[0]`, so a "chain" was never more than a single element. A dialog
can now configure an ordered list of fallback models in Chat Settings,
handed to `NewFailoverEinoChatModel` (sticky cursor plus a 30s
full-chain cooldown).
The list lives in the dialog's own `llm_setting.failover_llm_ids`, so no
new table is involved. A member that no longer resolves is skipped with
a warning rather than failing the turn.
Also removed: `tenant_model_group` / `tenant_model_group_mapping`, which
nothing ever read (the DAOs were constructed but never called, and no
frontend or Python code referenced the concept). Their removal takes an
explicit drop migration with it, plus the account-deletion cascade that
queried them.
### 3. A hung MiniMax stream (independent of the agentic work)
With any mode selected, a chat rendered its whole answer and then sat on
"thinking" forever. Root cause is `minimax.go:256`: MiniMax sends `data:
[DONE]` but leaves the HTTP connection open, and the code waited for the
scanner goroutine's EOF *after* `HandleStreamingResponse` had already
returned. That receive can only end when `streamCallTimeout` (20
minutes) expires.
Diagnosed by capturing a real SSE stream (the complete answer arrives,
the terminal `final: true` never does) and a goroutine dump (6 requests
parked in `chan receive`).
## Two review findings fixed on the way through
- **KB-scope authorization**: the agentic branch bypassed quote
resolution, and an empty KB scope made `buildBoolQueryFromCondition`
drop the `kb_id` filter — so a citation could resolve a chunk belonging
to a different KB in the same tenant. The agentic branch now requires a
non-empty scope and otherwise falls through to the regular path.
- **Stale documentation**: `agentic-rag-failover-groups.md` described
the "automatically include every tenant model" strategy that upstream
had already removed. It was rewritten for the per-dialog scope and then
dropped entirely, since the design now lives in the code it describes.
## Verification
- `bash build.sh --test`: `admin`, `dao`, `service`, `service/dataset`
and `entity/models` all pass
- The MiniMax fix was verified end-to-end against a live server: before,
the turn hung indefinitely; after, it completes in **1.9s** with `final:
true` present
- Frontend: 9 tests added; type-check and lint clean on the touched
files
## Not included
- **Attachment support in agentic mode.** Text attachments could be
appended safely, but images have no safe fix: the agent's toolset is
built around corpus retrieval and has no image input channel. Fixing
only the text path would leave the feature half-supported and harder to
diagnose than now. Planned as a follow-up PR, with the design synced
here first.
- Tool-calling is not enforced as a group constraint. `is_tools` is a
provider-declared flag rather than a measured capability (187 of 659
chat models do not declare it), so gating on it would reject working
configurations while admitting broken ones.
### Overview
**1. Python harness paths ported to Go**
- Table chunks render as Markdown.
**2. Field-driven `metadata_search`**
- Was pinned to enum `["title"]`; now a per-session catalog (parser_config declarations + observed keys/values) drives tool schema and seed.
- System keys blacklisted; falls back to old spec on empty catalog/no resolver.
- `graph_fanout` skips paid extraction when no field is usable.
- Harness honors dialog budget (`top_n`, `similarity_threshold`, `rerank_candidates_count`); fixes `top_n` being overridden by `DefaultTopN=12`.
**3. `metadata_search` is a selector, not a retriever**
**4. Outer-loop routing**
- Previously treated no-tool-call replies as final, letting KB questions be answered without research. `rag` and `summarize_document` now carry WHEN TO CALL / DO NOT CALL: `rag` required for any KB question; `summarize_document` only reads a doc_id already held.
### Summary
Refactor thinking log of agentic log
Example
```
[Agentic RAG] Starting research in high mode (self-check on, question fan-out on).
[Formalize] Kept the question as asked: "曹操是谁？"
[Planner] Split the question into 2 sub-questions to research: "曹操 生平", "曹操 历史人物".
[Prefetch] Searching 3 opening queries up front.
[BM25 search] Searching by keyword for "曹操 生平 简介".
[BM25 search] Found 5 passages in 1 document for "曹操 生平 简介".
[Hybrid search] Searching by meaning and keyword for "曹操 生平 简介".
[Hybrid search] Found 5 passages in 1 document for "曹操 生平 简介".
[BM25 search] Searching by keyword for "曹操 出生 去世 年份".
[BM25 search] Found 5 passages in 1 document for "曹操 出生 去世 年份".
[Hybrid search] Searching by meaning and keyword for "曹操 出生 去世 年份".
[Hybrid search] Found 5 passages in 1 document for "曹操 出生 去世 年份".
[BM25 search] Searching by keyword for "曹操 历史人物 主要成就".
[BM25 search] Found 6 passages in 1 document for "曹操 历史人物 主要成就".
[Hybrid search] Searching by meaning and keyword for "曹操 历史人物 主要成就".
[Hybrid search] Found 6 passages in 1 document for "曹操 历史人物 主要成就".
[Prefetch] Added 6 new passages to the evidence pool.
[RAGAgent] Round 1 begins with 6 passages in the evidence pool; 177 seconds of research budget left.
[RAGAgent] The pooled evidence answers none of the 4 slots; opening 3 research sessions this round.
[Function tool] Running the navigate_tree tool with "曹操 出生地".
[Function tool] Running the navigate_tree tool with "曹操是谁".
[Function tool] The navigate_tree tool returned 1 result from 1 document for "曹操 出生地".
[Function tool] Running the retrieve tool with "曹操 出生地".
[Grep search] Searching for the exact words "曹操 出生地".
[BM25 search] Searching by keyword for "曹操 出生地".
[BM25 search] Found 5 passages in 1 document for "曹操 出生地".
[Grep search] Found 5 passages in 1 document for "曹操 出生地".
[Function tool] The retrieve tool returned 4 results for "曹操 出生地", already in the evidence pool.
[Function tool] Running the navigate_structure tool with "曹操 出生地".
[Function tool] The navigate_tree tool returned 1 result from 1 document for "曹操是谁".
[Function tool] Running the retrieve tool with "曹操是谁".
[Grep search] Searching for the exact words "曹操是谁".
[BM25 search] Searching by keyword for "曹操是谁".
[BM25 search] Found 5 passages in 1 document for "曹操是谁".
[Grep search] Found 5 passages in 1 document for "曹操是谁".
[Function tool] The retrieve tool returned 4 results for "曹操是谁", already in the evidence pool.
[Function tool] Running the navigate_structure tool with "曹操是谁".
[Function tool] The navigate_structure tool has no compiled structure to read for "曹操 出生地".
[Function tool] The navigate_structure tool has no compiled structure to read for "曹操是谁".
[Function tool] Running the navigate_tree tool with "曹操 出生年份".
[Function tool] The navigate_tree tool returned 1 result from 1 document for "曹操 出生年份".
[Function tool] Running the retrieve tool with "曹操 出生年份".
[Grep search] Searching for the exact words "曹操 出生年份".
[BM25 search] Searching by keyword for "曹操 出生年份".
[BM25 search] Found 5 passages in 1 document for "曹操 出生年份".
[Grep search] Found 5 passages in 1 document for "曹操 出生年份".
[Function tool] The retrieve tool returned 4 results for "曹操 出生年份", already in the evidence pool.
[Function tool] Running the navigate_structure tool with "曹操 出生年份".
[Function tool] The navigate_structure tool has no compiled structure to read for "曹操 出生年份".
[RAGAgent] Round 1 finished: 0 new passages added; the pool now holds 6 passages, and 1 slot still unresolved.
[Draft] Drafted an intermediate answer of 729 characters from 6 passages.
[SCA] Evidence check: SUFFICIENT (confidence 0.85, reviewed 6 of 6 passages).
[Finalize] Finalizing a complete answer from 6 passages.
```
xugangqiang
Jin Hai
Caobaining1
Wang Qi
Denis Ermakov
buua436
Zhichang Yu
jay77721
maoyifeng
Lynn
qinling0210
Sbaaoui Idriss
Yingfeng