Commit graph

29 commits

Author SHA1 Message Date
Yiorgis Gozadinos
e8f00fcff4
Make get_config the only configuration lookup
haiku.rag.config exported two configuration instances: the lazy _config
behind get_config/set_config, and Config, loaded at import time. Nothing
linked them, and eleven signatures captured Config as a default argument,
so set_config could not reach the factories, the client, the store or the
MCP server. reranking/base.py went further and snapshotted the configured
reranker name into a class attribute at import.

Config is removed. Internal defaults are config: AppConfig | None = None,
resolved through get_config() per call. RerankerBase._model is None and
CohereReranker takes its model name as an argument, like every other
reranker.

The suite patched attributes on Config while production read the instance
get_config() returns, a different object, so those patches were no-ops
waiting to happen. They now go through get_config().
2026-08-19 14:43:40 +03:00
Lawrence Akka
cdd1b99e6b
Merge branch 'main' into chunk-metadata 2026-08-18 14:15:56 +02:00
Lawrence Akka
4eb72d6a49 Ignore warning caused by logfire
See https://github.com/ggozad/haiku.rag/pull/551#issuecomment-5327750999
2026-08-18 14:11:57 +02:00
Yiorgis Gozadinos
8a4d488a72
Give the MCP server one client for its lifetime
All ten tool bodies opened their own `HaikuRAG`, so every tool call paid a
connection open and, on object storage, refetched the index the previous call had
just cached. The client is now opened once, lazily so that calling a tool
function directly still works, and eagerly from the lifespan so an unopenable
database fails startup instead of every call. Teardown clears the cached client
in a finally, since `_lifespan_manager` can be re-entered and would otherwise
hand out a closed connection, including when the close itself fails.

`delete_document` no longer opens its own connection with `skip_validation=True`.
Keeping it separate broke consistency once connections became long-lived: the
delete committed on one connection while reads served from another, which with a
30s consistency interval showed the deleted document as still present. A
connection always sees its own writes, so sharing one is what makes delete
visible to the next read.

So the server no longer opts out of embedding-config validation. Drift that
validation rejects now fails MCP startup, where before the server started and
only `delete_document` worked while every read returned empty. Same-dimension
identity drift still starts a read-only server, matching every other read verb.
Delete under drift is now a CLI operation; CLAUDE.md and the CHANGELOG record it.
2026-08-18 14:58:18 +03:00
Lawrence Akka
2839a4434a Test that chunk metadata survivies FastMCP's wire serialization 2026-08-18 12:55:04 +02:00
Yiorgis Gozadinos
1e8e5e9f6f
Cover MCP, ingester and converter error paths
Add tests for the MCP tools' degradation contracts, malformed WebDAV
multistatus bodies, dry-run poller sweeps including the circuit-open and
discover-failure paths, FS source scheme and symlink handling, docling-serve
zip parsing, and the remaining embedding and reranker helpers. Parametrize
_strip_etag.

Drop the misplaced pragma on the analyze handler, which sat on the return and
left the except uncovered. Add one on the FS symlink OSError guard, which
resolve(strict=False) absorbs for every real link.
2026-07-26 19:14:26 +03:00
Yiorgis Gozadinos
c716be9640
cover the mcp path with no images 2026-07-25 15:00:36 +03:00
Yiorgis Gozadinos
c62fd78c7a
Accept images on CLI ask/analyze and MCP tools 2026-07-25 10:35:25 +03:00
Yiorgis Gozadinos
7fac35d2af
rename embed_image_query to embed_image; run description check on text path 2026-05-05 16:15:47 +03:00
Yiorgis Gozadinos
35d5f4416e
fill in vLLM error-path and MCP image-query coverage 2026-05-05 09:39:20 +03:00
Yiorgis Gozadinos
65d9c74224
expose image-as-query through MCP and the CLI. 2026-05-04 11:46:45 +03:00
Yiorgis Gozadinos
83ea6347b6
Fix CI failures: update tests for pydantic-ai 1.70.0 and FastMCP API changes 2026-03-24 14:02:03 +02:00
Yiorgis Gozadinos
5474465610
Fix mcp test 2026-03-24 13:36:14 +02:00
Yiorgis Gozadinos
d1fdbdeb21
Refactor MCP to use proper models, test 2026-02-20 17:39:02 +02:00
Yiorgis Gozadinos
b64d9721bb
Mark with pragma no cover areas that are just wiring. Consolidate tests and bring coverage to 100% 2026-02-13 11:56:37 +02:00
Yiorgis Gozadinos
20414ed959
Additional tests 2026-02-06 12:07:21 +01:00
Yiorgis Gozadinos
57fd08d600
Revert "Synthesis steps now select only relevant citations instead of including all" 2026-01-28 16:00:27 +02:00
Yiorgis Gozadinos
8a587ec554
Synthesis steps now select only relevant citations instead of including all 2026-01-27 15:56:29 +02:00
Yiorgis Gozadinos
5e55c0df54
replace pyright with ty type checker 2026-01-19 15:50:14 +02:00
Yiorgis Gozadinos
7cc561d1db
Refactor to make agents a top-level module. Bring in the conversational agent from the app 2026-01-12 12:37:08 +02:00
Yiorgis Gozadinos
f0a7abd953
Remove deep qa graph, everything now uses the simplified research graph 2025-12-18 11:41:10 +02:00
Yiorgis Gozadinos
4da50873bd
Make search limit configurable 2025-12-09 11:46:43 +02:00
Yiorgis Gozadinos
d808c6c425
Simplify citations in qa & graph agents 2025-12-08 15:55:27 +02:00
Yiorgis Gozadinos
c69b934eac
Incorporate additional docling meta in search(), expand_context() and relevant qa/research agents 2025-12-08 15:54:56 +02:00
Yiorgis Gozadinos
032625b0bb
Refactor everything graph-related under the graph module 2025-11-13 13:22:03 +02:00
Yiorgis Gozadinos
782fbafb3d
Make graph settings part of config for both deep ask and research graphs 2025-11-06 14:31:22 +02:00
Yiorgis Gozadinos
9075dac23a
Remove graph common and move its content to deep QA, research. 2025-11-06 11:02:32 +02:00
Yiorgis Gozadinos
1f22f74e1a
Add filtering to document listing in CLI/client 2025-11-04 16:02:58 +02:00
Yiorgis Gozadinos
43dac018c0
Test mcp tools 2025-10-08 09:20:26 +03:00