Document.source names the configured database, as SearchResult and
Citation already do. A listing spanning databases is unreadable
without it, and `--database NAME list` could not name the one it
opened.
Reciprocal rank fusion compares ranks, so every database contributes its
own best matches whether or not they answer the question, and results from
databases holding nothing relevant displace better ones. Measured on one
corpus split three ways over 3,045 queries: retrieval MAP 0.6044 without a
reranker against 0.9798 for the same corpus in a single database, and
0.9918 against 0.9914 with one.
Chat answers with the same capabilities `ask` does, so it federates as
naturally as `ask` and `analyze` — but it went through the one-database
guard and refused a configured set outright, which left no way to chat
across several databases.
The guard was the visible half. `run_chat` also defaulted `db_path` to the
single default path whenever it was None, so lifting the refusal alone
would still have opened one database. It now leaves the path unresolved
when `lancedb.databases` names the set, and the client resolves it.
Listing and counting documents fan out over the set, which is what the
document filter reads, and visual grounding resolves the database holding
the cited chunk through the citation's source: chunks, pages and bounding
boxes all come from that one database. A limit on a listing means that
many documents in total, not that many per database.
The info modal reports every database it covers, each under its
configured name and without its location, since names are the only
identity that leaves the configuration. `database_lines` is what one
database reports about itself, shared by both paths, and it reports a
failure as a line so one unreachable database does not cost the report on
the others.
`inspect` stays a one-database command. It browses one database's
documents and chunks, so a set has nothing to show it.
lancedb.databases, the sources argument and the --database selector had no
documentation. Adds a Several Databases section to the storage
configuration page covering the name-to-location map, its mutual
exclusion with uri, the shared embedding configuration the set requires,
and which commands cover the set against which work on one. Adds a
Searching Several Databases section to the Python API page, --database to
the CLI's global options, the key to the sample configuration, and one
README feature line.
`Agent.from_spec` raised `TypeError` on `RAGCapability` and `AnalysisCapability`,
whose constructors take a state class, packaged instruction text and a tool-name
set, and silently omitted both from the generated spec schema. The two
zero-configuration capabilities constructed but with `id=None`, so pydantic-ai's
duplicate-id rejection no longer held and a spec could register two citation
policies, defeating the single-decision-maker invariant.
Override `from_spec` on all four, delegating to `create_capability()` so ids and
instructions come from one place. The spec surface is `db_path`, `config`,
`defer_loading`, `request_limit` and `vision`; a live `HaikuRAG` client stays out
of it, and a `config` mapping is validated through `AppConfig`.
Sixty-three comments said what the next statement already said: # Connect to
LanceDB above connect_lancedb, # Path object above isinstance(source, Path),
# Get page numbers from provenance above the prov loop, # Clear and populate
results above list_view.clear(). They cost a read and carry nothing.
The line is whether a comment restates one statement or labels a phase. Phase
labels stay: the migrations keep # Create staging table with new schema and
# Copy from staging to final table in batches, each heading ten lines of a
long procedure. So do comments carrying a fact the code cannot: the
merge_insert update-only note on document_meta, why the poller builds sources
eagerly, why create_document_from_source returns a list for directories, that
indexes need training data, the field-group markers in the config models, and
the file:// URL-encoding note in create_document_from_source.
capabilities/ is untouched. Its docstrings sit next to prompt surface, and
changing them needs an eval to back it.
The cassette-recording docs were wrong three ways. They named
tests/test_qa.py::test_qa_anthropic, which no longer exists; they targeted
whole modules, so a rewrite would re-record cassettes for services the
recorder is not running; and they used COHERE_API_KEY where the SDK reads
CO_API_KEY. docs/development.md now names exact tests with -n0, and the keyed
example is test_cohere_reranker, which owns the one cassette recording
api.cohere.com.
overview.md repeated the landing page: the same install-and-ask block and
five of six identical links. It was positioning prose, where the docs had no
page describing how the system works.
Rewrite it as Architecture, following the data through: source adapter,
converter, chunker, embedder, transaction; then storage and its versioning;
then retrieval, with the 10x rerank fetch and section-bounded expansion; then
the two capabilities; then laptop versus ingester. Retitled in the nav and on
the landing page, filename kept so existing links resolve.
Extras were listed in three places and none was complete.
docs/installation.md now carries a table of all fifteen slim extras, what each
provides, and which the full package already includes.
haiku_rag_slim/README.md names them and links there. The claim that other
providers need their own pydantic-ai extra was wrong: haiku.rag-slim defines
anthropic, google, groq, mistral, bedrock and vertexai itself.
configuration/storage.md opens with the four operational constraints, which
were either buried in an S3 section or undocumented: one writer per URI,
reader lag by read_consistency_interval_seconds, migrate after a
schema-changing upgrade, and the fixed embedding dimension with what
ConfigMismatchError means and which rebuild mode resolves it.
The one-writer rule is stated as a haiku.rag constraint, which is what it is:
the multi-table lock, version snapshot and rollback are process-local, so a
second writer can commit inside another's transaction and be reverted by its
rollback. storage.md and ingester.md both claimed it was a LanceDB property
that corrupts manifests. The S3 deployment section now links to the
constraint instead of restating it.
Get started reads index, Quickstart, Installation, Architecture. The landing
page's list was missing Installation.
haiku.rag.ingester.sources was never ingester-only: one-shot client
ingestion resolves adapters through it (create_document_from_source), and
convert() now fetches through HTTPSource, so the core client imported into
the ingester package to reach them.
Move the package to haiku.rag.sources and update every import. No shims:
haiku.rag.ingester.sources is gone.
The haiku.rag.sources plugin entry-point group is unchanged, so third-party
source packages need no edit — the group name now matches the module path it
always implied.
Source unit tests move to tests/sources/. test_source_plugins.py stays in
tests/ingester/: it drives a PeriodicPoller against the job repo, so it is
plugin wiring through ingester machinery rather than a source test.
search.limit was documented as 10 in three places while the default is 5.
The documented way to disable reranking, provider: "", is a valid
ModelConfig, so it raised "Unknown reranking provider" — disabling means
omitting reranking.model or setting it to null. The inline provider list
named four of the six rerankers. prompts.picture_description: null fails
validation, since the field is a non-optional str.
storage.data_dir: "" coerced to Path("") — the working directory — while two
doc pages promise the platform default and soliplex's example config relies
on it. Empty or whitespace now resolves to the platform directory; an
explicit "." is still honoured, so a config that wants the working directory
says so.
Three tests keep this from drifting again: every fenced yaml block in the
docs validates against AppConfig, every value in the complete example either
equals its default or is listed as a deliberate deviation, and empty
data_dir resolves to the platform default.
init-config's test reimplemented the command body instead of invoking it,
which is why the command carried a coverage pragma. It now goes through
CliRunner, with the refuse-to-overwrite guard covered too.
Every section inherited plain BaseModel, so unknown keys were dropped
silently: providers.docling_serve.timeout was documented for months while
being ignored, and a typo in any setting took the default. Sections now
derive from ConfigModel, which forbids extras, so a stale or misspelled key
fails with its path. This already found search.context_radius in a live app
config and providers.vllm in soliplex's example.
converter, chunker and chunker_type are Literals. Sizes, limits,
dimensions, token budgets, attempt counts and breaker thresholds must be
positive; retention, delays, intervals and cooldowns non-negative;
similarity_threshold within 0-1; port within 0-65535. port 0 keeps its
OS-assigned meaning and worker_count allows 0 for an API-and-reaper-only
process.
get_reranker caught ImportError and returned None, so a configured reranker
whose extra was missing silently disappeared. It now propagates.
raise_missing_extra names the install command and re-raises when the failure
came from inside an installed package, so a broken transitive import is not
reported as a missing one. zeroentropy imported bare and now guards like the
others.
The haiku.rag package declares the jina extra. jina-local already worked
there through cross-encoder's transitive transformers and torch; the
resolved package set is unchanged, but the support is now promised rather
than inherited.
Provider fields stay unconstrained: get_model ends in a pass-through to
pydantic-ai for any provider it supports, so a Literal there would reject
valid configurations.
The setting was documented but did not exist on DoclingServeConfig, so it
was silently dropped, and DoclingServeClient.from_config never forwarded
the timeout parameter it already accepted. The per-request timeout was
therefore pinned at the constructor default of 300s with no way to change
it. Add the field, forward it, and reject a non-positive value.
Also parametrize over the checked-in *.yaml.example files and validate
each through AppConfig, so an example that no longer loads fails a test
rather than a user's first run.
Every `Store` built its own connection with its own caches and discarded them on
close, so the index a vector query loads was refetched by the next connection.
On object storage that first fetch dominates: measured on a ~500k-chunk 2560-dim
corpus over a ~200ms link, the first query cost ~41s and the second ~3s, and a
new connection reusing the session cost ~7s instead of ~47s.
`connect_lancedb` now passes a process-wide session, keyed on the configured
cache sizes so a caller asking for different sizes gets its own.
Also sets `read_consistency_interval`, defaulting to 30s. It was None, meaning a
connection never re-checked for other processes' writes. Per-call connections hid
that; a shared session makes connections long-lived enough for a reader to go
stale against the ingester.
All three settings reject negatives at the config boundary. A negative cache size
raises OverflowError and a negative interval panics inside Lance, so neither is
catchable further in. Zero stays valid for both: no cache, and check on every
read.
The routing tests now assert the kwargs they care about rather than the full call
signature, since every connection carries the two new kwargs.
Current results moved above Methodology and Running Evaluations. A reader arriving from an external link met four screens of CLI flags and download instructions before any number.
Benchmarks is promoted to a top-level nav entry, out of Reference, which holds Development and Changelog.
The page move itself changes no wording or figures.
IBM's MTRAG benchmark (ClapNQ domain, pinned repo SHA): retrieval with
Recall@k/nDCG@k against binary qrels, gold-prefix QA replaying reference
conversation prefixes as message history, and live-session replay
carrying the model's own answers and tool history across turns.
Corpus population gains a bounded, resumable batched ingest path.
ConversationInput case type with transcript rendering for the judge,
eligibility-aware citation scoring, refusal precision/recall via a
label-aware RefusalJudge, per-turn verdicts with judged-turn coverage,
and per-turn tool-traffic attributes counted from each turn's new
messages so the arrays survive prior-turn compaction.
No `DatasetSpec` declared `search_filter`, so `resolve_search_filter`
and the `--filter ""` clearing rule reconciled the flag against a
default that never existed. The flag alone covers the case. An empty
clause reaches `ChunkRepository.search`, which already treats it as
unfiltered.
Rename to `document_filter` throughout, matching
`run_capability_question`'s parameter and the metadata key that lands
in Logfire.
`_stub_spec` merges its overrides, so a test can override a loader
instead of rebuilding the whole spec.
The `--filter` examples used `uri LIKE '%arxiv%'`, which matches no
`orb_text` document: its URIs are bare arXiv ids such as `2407.01528v3`.
A clause that matches nothing scores MAP 0 rather than erroring, so the
example failed silently.
`await_args` is typed `_Call | None`, so subscripting it fails
`ty check`; `call_args` carries the same call for an AsyncMock.
Both optional capabilities read what earlier questions retrieved and cited from
the capability's state, so a host that carries only the message history hands
every run an empty record. Compaction then replaced the earlier evidence with
receipts and retained nothing, and the loss was invisible: the citations the host
already displayed were still there. It now refuses when it finds evidence from an
earlier question and no record of what that question cited.
`state_carried` reaches the optional capabilities through discovery, so the
refusal distinguishes a host that never carries state from a question that simply
cited nothing.
The documentation taught the pattern that breaks: the compose example is now
stateful and the requirement is stated where each capability is introduced.
The app's browser storage was doing exactly this, keeping only the fields the UI
reads. It now persists the whole namespace map, so the citation policy's
violations survive a reload as well as the evidence record.
The README feature list and the overview stopped at the analysis capability.
Both examples and the app backend composed agents without the capabilities the
documentation recommends alongside an evidence capability.
custom_agent.py ran each input as an independent agent run, so it needed a state
dict and a carried history before compaction could mean anything there: without
state the evidence record is empty, and earlier evidence would reduce to
receipts retaining nothing.
The documentation claimed a conversation that has never cited anything is not enforced,
which reads as though a first question could answer from fresh evidence without
declaring it. Enforcement needs neither condition to hold: no evidence outcome in this
question, and nothing cited earlier.
The structured-output test accepted a redirect or a recorded violation, so a break in
ending detection would still have passed on the backstop alone. The cite tool is
available in that scenario, so it asserts the redirect.
An output tool call is a `ToolCallPart` like any other, and treating every tool call as
intermediate meant a model could search, skip citing, emit its structured answer and
finish with neither a redirect nor a record. A response ends the question when it
carries no tool calls, or when one of its calls names an output tool.
Some endings are not visible from a single response — a host running
`end_strategy="early"` can finish on text beside a function call — so `after_run` is
the backstop: it cannot ask the model for anything by then, but it records a question
that reached the end of its run undeclared. That also covers a question that was asked
once, ignored, and finished anyway, which previously returned early on the redirect
marker and went unrecorded.
The capability documentation and the changelog said a question that gathered no
evidence is left alone. That describes neither the code nor the intent: enforcement
applies wherever there is something to declare, which includes a follow-up that reuses
evidence cited earlier without searching again.
`CitationPolicyCapability` makes the single enforcement decision, whatever mix of
evidence capabilities is registered: two of them must not each demand a citation for
one answer. It decides in `after_model_request`, when a response carries no tool calls
and the question can still be redirected. An explicitly ungrounded answer is a
declaration and is left alone; a question that gathered no evidence is left alone too,
read from the ledger rather than from a searches dict that a new question clears. When
the cite tool is already withdrawn the question is recorded in
`CitationPolicyState.violations` instead of pointing the model at a tool that is gone.
Registering it is the only switch. `DiscoveredEvidence` and discovery move to
`capabilities.evidence` so both optional capabilities share them, and `cite_available`
joins `evidence_tool_names` as public for the same reason.
Measured on Qwen3.6-35B over two arms of 37 questions, 29 of them unanswerable from
the corpus: explicit ungrounded declarations rose from 23 to 26, grounded answers to
unanswerable questions fell from 4 to 2, answerable questions stayed at 8 of 8, and
one redirect fired in the whole arm. Reading every answer found no invented grounding.
The nemotron rows for retrieval and for both capabilities are re-measured on
this release with no reranker, judged by Qwen3.6-35B with thinking on. Rows for
other embedders and older versions keep their own attribution.
The capability pages said tool results from earlier turns are replaced before every
model request. That is now the compaction capability's job, and only when a host
registers it, so both pages point at it instead of describing it as automatic.
`RAGState` gains its `evidence` field, and the note about per-run resets now says
what a resumption keeps.
`_compact_old_tool_returns`, `PRIOR_TURN_NOTICE` and `turn_start` leave
`RAGCapabilityBase`, along with its `wrap_model_request` hook. The evidence
capabilities now retrieve and validate, and nothing else. Registering the compaction
capability is what rewrites a request; leaving it out sends the transcript untouched,
which was never a choice a host could make before.
The boundary is the recorded question identity rather than message shape, so a
resumption compacts what lies below the question in progress instead of switching
compaction off for the whole run. The newest earlier evidence return carries the
capsule and every other becomes a receipt, so one capsule exists by construction and
every return stays paired with its call.
Pictures of cited evidence are fetched through the capability that retrieved them and
re-attached beside the capsule with fresh labels. Ownership of a picture on the wire
requires the machine tag we write and an image directly after it, since neither
position nor prose is proof: several tools' results can arrive in one request, and a
user quoting our wording above their own picture had it removed. A picture that cannot
be fetched or decoded is emitted with neither its image nor its label.
The chat TUI and the example backend register the compactor, being multi-turn.
`client.ask`, `client.analyze` and the MCP tools do not: a single-shot question has
nothing earlier to compact.
The request-limit notice said only the cite tool remained available, but
chat registers rag and analysis in one agent, so exhausting analysis
claimed rag_search was gone too. Scoped to the capability's own tools.
The cite window was counted over every model request once loaded, so
turns spent on another capability expired it before the model was ever
placed where citing was the obvious move. Count only requests whose
preceding response called one of this capability's tools; engagement is
also the only thing that can loop, which is all the bound guards against.
Also: _count_tool_traffic returns a named tuple rather than four bare
ints, and counts failures only for this capability's tools, so host-tool
retries and output-validation retries no longer read as its failures.
