Master polished two Chinese sentences this branch also edits — an
expectation is now 陈旧 rather than 过期, and the package-root sentence
spells out 包（package）. Taking master's wording alongside this branch's
own additions leaves the recorded pair fingerprints stale, so they are
re-recorded against the merged text.
llm.discoverModels was reachable from any declared trusted host. The
method takes a caller-supplied baseURL and makes the host issue a GET to
it, then reports the status or the parsed body — so on a LAN deployment
an anonymous caller had a probe for whatever the host can reach and the
browser cannot, plus a path that carries a draft credential. The
PRIVILEGED_METHODS doc already states the rule this broke: trustedHosts
is a DNS-rebinding fence, not authentication, so the configuration plane
stays loopback-same-origin. It is in that set now, asserted both against
the hand-built fence and over real HTTP beside the catalog reads that
deliberately stay reachable.
supportsDiscovery and listModelDiscoveryNamespaces are gone. The field
was required on the wire and read by nobody: its own contract said a
surface should offer the action "instead of naming an adapter family it
would have to hardcode", while the surface hardcodes llm-pi-ai in two
places and gates the button on whether there is anything to probe. Its
shape did not fit the second caller either — the create card has no row
to read a per-row field from. Keeping a required field alive for a
consumer that may never arrive costs every producer and fixture a value
nobody consults, which is exactly how the fixtures drifted. The registry
that fed it had no other production consumer, so registration and
disposal are now observed through the offer itself.
The Agent Note claimed the key is never logged, which the wire schema
beside it already contradicts, and predated both the provider field and
the catalog-answer path. The two new public types pointed at core.md
without a type-equiv block or manifest entry, so the generated service
catalog named documentation that did not exist.
Clicking "fetch available models" on a built-in provider went to the
network. That is the wrong source: pi-ai's registry is the authoritative
list for its own providers, and it carries the context windows and output
caps a `GET /models` listing does not disclose. Asking api.deepseek.com
what DeepSeek serves is both slower and worse, and against an endpoint
that answers a different shape it failed outright.
Interrogation is still keyed by settings namespace — the provider being
added has no route — but the request may now name the route it is
editing. An adapter that already describes that route answers from what
it knows, needs no endpoint at all, and never touches the network; only a
route the catalog does not describe reaches the wire, and one naming no
endpoint is told to set one or enter its models by hand.
`ConfigurableProviderView` gained `supportsDiscovery` so a surface offers
the action where a namespace can answer instead of hardcoding an adapter
family.
Three narrower corrections ride along. Discovery no longer claims Azure
or Codex: Azure authenticates with an `api-key` header and an
`api-version` query despite its OpenAI lineage, and Codex uses OAuth, so
both reported an authentication failure as a provider with no models.
Cancellation during the body read escaped as the raw abort reason rather
than a coded ABORTED. And the schema comment claiming the probe key is
never logged overstated it: the host neither stores nor returns it, but
it rides the client's outgoing envelope like every other secret-bearing
payload, and redacting that tap is a configuration-plane-wide change.
Once a pi-ai route became a declaration rather than a catalog lookup,
adding an OpenAI-compatible gateway meant knowing its model ids up
front. Most such endpoints publish that list at `GET /models`, but no
seam operation could ask: every one is keyed by a registered provider
route, and the provider being added has no route, no stored profile,
and no stored credential — the endpoint and key are values in a form.
Interrogation is therefore keyed by settings namespace, which a
configuration surface already holds from the configurable-provider
directory. `registerModelDiscovery` offers it per namespace,
`discoverModels` asks, and the request carries the draft itself. The
reply is candidates, not a catalog: every field but the id is optional
because most listings disclose nothing else, and adopting one is a
settings write like any other. Nothing here reads or writes settings or
credentials, so `settings.yaml` still decides what a route serves.
`llm.discoverModels` carries the same draft over the wire. Its apiKey is
the third and last payload a secret may ride, and it is never stored,
logged, or echoed; every refusal folds into `model-discovery-failed`,
naming the endpoint asked but never the credential offered.
The pi-ai side is a plain GET for OpenAI-compatible protocols only —
their listing shape is the one gateways, self-hosted servers, and the
official endpoints agree on. Others say so, sending the user to
hand-entry rather than reporting a guessed shape as an empty provider.
The reply is read under a four-megabyte ceiling held on the bytes
actually received, because the endpoint is a URL the user typed.
A documentation rescan on master moved this README's Testing section out
of the package. This branch was still editing that section, so the
rebase asked which structure wins; master's does, and the branch keeps
only the Catalog resolution section its own change adds.
Re-records the pair fingerprint against the merged text.
pi-ai resolves a request's apiKey override only through a provider that
declares an api-key method: resolveProviderAuth short-circuits to that
method when the override is present, and otherwise falls through to the
credential store and then to ambient discovery. A provider with no
api-key method at all therefore resolves to nothing, and the request
fails with "Provider is not configured" before any network I/O.
Two routes hit that. openai-codex ships OAuth alone, so moving off the
/compat dispatch broke a profile that names a key for it — the old path
handed the token straight to the provider. And a catalog route naming an
api was being rebuilt with the harness's own auth, so `openai: {api:
openai-completions}` stopped reading OPENAI_API_KEY, contradicting the
documented promise that omitting a credential keeps provider-native
discovery.
Auth is now one decision for both constructions. A catalog route keeps
its installed provider's auth, through an api override too: which
environment a provider reads belongs to the provider, not to the wire
format its models speak. A catalog provider with no api-key method gets
the harness method beside its own, but only when the profile names a
credential — a keyless codex profile keeps the honest refusal, since
this adapter holds no OAuth store to resolve through.
Materialization now spreads the installed entry instead of enumerating
the result, so a Model field this package does not model survives a
pi-ai upgrade; headers went missing from an nvidia route exactly that
way once already. providerInfo reports the configured displayName, which
also joins the registration facts so a rename re-registers rather than
leaving the old label in every selector. A refused registration swap
gets its own diagnostic naming the route, matching the directory swap
beside it.
The README documented endpoint interrogation this layer does not
implement, and still described unknown providers as kept-last-good after
they became legal declarations refused at the write point. The Agent
Note claimed per-model reasoning configurability the schema never had,
required capacities the route now defaults, and stated an apiKey
override that short-circuits unconditionally.
The adapter omits the seam's reasoning field whenever a model carries no
reasoning metadata, which is the model's own property and says nothing
about where the model came from. Both the JSDoc and the Agent Note read
as though only hand-declared models were meant, so a reader would infer
that the 251 installed-catalog models pi-ai marks as non-reasoning still
offer their single off level. They do not, and that is the point: a
picker holding only off misrepresents a provider that thinks by default,
because off dispatches the same bytes as naming no effort at all.
Behavior is unchanged; only the prose that describes it was narrower
than the contract. adapter.spec.ts already pins the catalog case through
openai/gpt-4.1 and catalog.spec.ts pins the hand-declared one.
Three defects surfaced while driving the Models page.
A hand-declared model needed an explicit contextWindow and maxTokens,
but a provider listing usually returns ids and nothing else — so the
page happily wrote a profile the adapter then rejected, which took the
whole namespace down silently. Capacities now fall back to the route's
`defaultContextWindow` (262,144) and `defaultMaxTokens` (32,768). Both
are guesses by construction, which is why they are route fields a
deployment corrects once rather than constants buried in the adapter;
the fallback sizes the model and never becomes a per-request cap.
That silent failure was the second defect. A schema-valid profile the
adapter could not serve was stored and only rejected later, disabling
every route in the namespace with nothing said. `dsh-settings` gains an
optional `validate` on registration — a check for what a schema cannot
express — and `llm-pi-ai` refuses an unserviceable section at the write
that produced it. A stored section that fails keeps the namespace's last
good value, as a schema failure already did, so an externally edited
document still cannot strand the owner. The plugin's own last-good
fallback goes with it: nothing reaching it can fail any more.
Third, a model with no reasoning metadata advertised the single level
`off`, which pi-ai translates to *omitting* the reasoning option — the
same request naming no effort produces. Selecting it disabled nothing,
so a provider whose default is to think kept thinking with `off` shown
as selected. Such a model now reports no reasoning capability at all,
which is the seam's way of saying the control is unavailable, and the
per-model `reasoning` flag is gone: without a thinkingLevelMap to spell
levels it could only invent them.
The protocol table narrows to the three a hand-declared route reaches
today, most-reached first so a surface offering a choice defaults to the
one gateways actually speak.
Review found four defects in the declared-provider work.
`PiAiAdapter` reused one `Models` collection and mutated it whenever the
configuration changed. `Models.streamSimple()` resolves its provider
lazily — when the stream is first consumed, which is after the adapter
awaits the route's credential — so a configuration change landing in
that window let an in-flight request finish under a configuration it
never resolved against, or fail on a provider that no longer existed.
Each resolution now produces an immutable snapshot and every operation
captures one before its first await, which is what makes the seam's
per-step freeze (`llm.prepareCall()`) hold end to end: switching models
mid-reply takes effect on the next step, never inside the one in flight.
`defaultMaxTokens` was materialized from the catalog's `Model.maxTokens`.
The two answer different questions: pi-ai requires that field as the
model's output capability, while the seam's is a cap the deployment
chose to send on requests naming none, so every request had started
carrying a number nobody picked. Only an explicitly configured cap
reaches the seam now.
The configurable-provider directory was refreshed by disposing its
registration and making a new one. A candidate set the registry refuses
— a profile keyed `deepseek-official`, which llm-deepseek declares —
left the whole directory withdrawn and the Models page empty, silently,
because the settings callback contains the failure. The seam's
registration handle now carries `replace()` with the same
validate-first atomicity `registerAdapter` has.
The protocol table offered every pi-ai streaming API, including four
whose authentication a profile cannot express: Bedrock signs with SigV4
over AWS credentials and a region, Vertex needs a project, a location,
and ADC, Azure needs provider environment plus an api-version, and Codex
uses OAuth. Offering them handed back routes that cannot authenticate.
Catalog routes still reach them through their own provider.
A pi-ai route had to name an installed catalog provider, served that
catalog's models verbatim, and could override only the endpoint. An
OpenAI-compatible gateway, a self-hosted server, or a model newer than
the pinned pi-ai release was therefore unreachable, and a stale context
window could not be corrected without upgrading the package.
A route is now a declaration whose defaults come from the installed
catalog. `catalog.ts` merges that catalog under the profile's own model
entries, `provider.ts` builds the pi-ai Provider (reusing the catalog
provider when the route keeps its protocol, so implementations this
package cannot reconstruct keep working), and the adapter serves every
operation from one `createModels()` collection. That also retires the
`@earendil-works/pi-ai/compat` import, which pi-ai documents as a
temporary entry point it deletes with its ModelManager migration.
Credentials stay on the harness seam: the resolved key rides the request
as pi-ai's highest-priority auth override, so `Models` holds no
credential store and a named-but-missing reference still fails loud
instead of falling back to an unrelated ambient key.
A model's configured maxTokens now reaches the seam as defaultMaxTokens.
A catalog entry may now carry its own `maxTokens`, matching the shape the
pi-ai adapter already exposes. Exact-model resolution prefers it over the
profile value, so capping one model no longer means capping the route.
Defaults are unchanged: an entry without a cap, and any unlisted
pass-through id, still resolve to the profile `maxTokens` (256,000), and the
shipped catalog keeps its context windows.
Two lifecycle holes the registry and the consumer helper left open.
`AdapterRegistrationHandle.replace` had no liveness guard: after the handle's
disposer ran, a replace put routes back into the registry with nothing left to
release them, so the adapter leaked permanently. `owned` being empty cannot
carry that fact, because `replace([])` is the legal empty-section state, so the
disposer records it explicitly.
`installSettingsSection`'s watcher lacked the guard its own disposer carries:
a stored change landing while the consumer unloads reached `onChange`, which
re-registers routes against a fiber whose resources are being released.
Also documents `withFileLock` in the atomic-write README (it claimed one
export), records the age-based lock takeover as a known limitation, and lists
ctx.settings and ctx.credentials in the architecture capability table.
The remaining P1 from the #939 review, plus the P2 it shares a mechanism with.
Nothing carried a version, so two tabs editing one namespace silently
overwrote each other — reproduced as tab B's `reasoning` lost to tab A's
older draft. The seam's per-namespace write queue orders writes; it cannot
tell a fresh writer from one replaying a snapshot a predecessor superseded.
Each namespace now carries a monotonic `revision` over its RAW section. A
write may send `expectedRevision`, checked at the FRONT of the queue (not at
call time, which would race the very predecessor it guards against); a
mismatch rejects with `SettingsConflictError` → `settings-conflict` on the
wire, carrying both revisions. The editor captures the revision it opened at
and, on conflict, asks the user to reopen rather than replaying its snapshot.
The same counter fixes the missing broadcast. `settings/updated` is gated on
the resolved value — correct for consumers, wrong for configuration surfaces:
storing an override equal to the composition base leaves the resolved value
alone while changing what the document says (the field is now overridden, not
inherited) and moving every open editor's revision. `settings/document-updated
(ns, revision)` fires on any raw-section change, in-process or external, and
`host/settings-changed` now rides it.
That event also closes the stale model picker: editing a provider's `models`
changes no route, so `llm/adapters-updated` never fired and an open picker
kept serving the old catalog. A change to an exposed provider namespace now
emits `host/models-changed` too — that namespace holds the catalog.
Docs: both sides of the five touched README pairs, a type-equiv block for
`SettingsPathOp`, and an Agent Note recording what the plane exposes and who
may overwrite what. The deferred wire-redaction gaps (secrets behind
union/intersection/transform, `.default(...)` in the served envelope, schema
text in rejection messages, `new Function` rehydration, pi-ai's `headers`) are
recorded as TODO(settings-wire-redaction) and in Known Limitations rather than
half-fixed.
Five findings from the #939 review, each reproduced before being fixed.
**Configuration reads are as privileged as writes.** `settings.describe`
returns every exposed namespace's configuration and `credentials.describe`
reports whether an arbitrary environment-variable name is configured and from
where — reconnaissance no anonymous caller should have. Both join
PRIVILEGED_METHODS, so the whole configuration plane is loopback-only until
real authentication exists; `trustedHosts` was never authentication. The model
catalog stays reachable: it carries no endpoints or key state, and a LAN
client's model picker legitimately needs it. Asserted over a real HTTP server,
because the Host header a browser actually sends is what decides this.
**The proxy serves only namespaces a registered model provider addresses.**
The settings seam is general — any plugin may register one — but the Web
configuration plane is the model-provider surface. Without the gate, every
future `settings.register()` would silently become remotely readable and
writable configuration. An unregistered namespace and an unexposed one answer
identically, so no caller can enumerate the registry one probe at a time.
**Path-addressed writes replace the redacted-document rebuild.** The editor
reads the REDACTED descriptor, so rebuilding a section from it and replacing
wholesale deleted every literal secret the wire never returned — reproduced as
`{baseURL, reasoning}` in, stored `apiKey` gone out. `settings.mutate` applies
set/unset ops to the section as it stands at the front of the seam's write
queue, and the client names only fields it can see, so an unseen secret is
untouched by construction rather than by care.
P2s in the same pass: `llm/adapters-updated` now contains async listener
rejections (an uncontained one escaped as unhandledRejection, contradicting
the documented "observer failures are contained"); llm-deepseek's retry-policy
swap uses the atomic `registration.replace` instead of dispose-then-register,
which published `[]` then `["deepseek-official"]` so an observer saw the
provider disappear and come back; and a transport rejection no longer strands
the page in `loading` or a card in `busy`, with removal failures surfaced on
the page banner instead of swallowed.
Both provider READMEs state what actually holds: credentials-local now
documents the physical-line editor, the read-modify-write under the
writer lock, and a Security boundary section saying plainly that the file
mode stops other OS users and not the model. sandbox-policy documents
readDenyPaths and its per-backend enforcement. The llm READMEs carry the
registration handle, pi-ai's credential-miss semantics, and DeepSeek's
same-generation snapshot; app-boot and the CLI README stop describing
$DSH_HOME/.env as an environment layer.
A new Agent Note records the round (and the prior seam note cross-links
it); the sandbox and core catalog pages gain readDenyPaths and
AdapterRegistrationHandle with their manifest entries. The headless
missing-credential snapshot re-records for the reworded guidance, pi-ai
gains the Loader-composition guard its twin already had, and the
deliberate provider symmetry is marked for the clone detector.
The policy home's resolve() now stamps readDenyPaths, so every consumer
that pins the resolved shape (bash-sandbox hand-off, tool-fs stamps)
carries it, and three uncovered branches gained real tests: landlock
reporting partial enforcement for a denial it cannot express, the
policy's default under programmatic construction, and both ambient
credential paths in llm-deepseek without a mounted seam.
Four review findings across the seam and both adapters.
registerAdapter now returns a handle carrying replace(providers): the
candidate route set is validated in full before anything moves, so a
route another adapter owns leaves the previous registration intact, and
the swap itself is one synchronous section with no observable gap. pi-ai
uses it instead of dispose-then-register — the old shape dropped every
route when the new set conflicted, and its facts cache could then equal
the registry's, so reverting to a working configuration never re-applied.
Its registration facts are also sorted by provider, so a settings
document that merely reorders keys no longer triggers a swap.
DeepSeek's per-request snapshot now carries the credential facts, and
resolveApiKey receives it instead of re-reading the raw config: a
settings generation the resolver rejects can no longer contribute its
literal key to a request the previous generation's endpoint serves.
pi-ai only defers to the SDK's provider-native discovery when a profile
names no credential at all; a configured apiKeyEnv that misses now fails
with MISSING_CREDENTIAL naming the route and the reference, instead of
handing pi-ai undefined and letting it authenticate with an unrelated
ambient key.
The eager boot-time credential probe is gone: it could run before the
credentials service mounted and reported every failure as a missing key.
The route stays registered and browsable; the first request gives the
accurate error, whose guidance now leads with the credential store and
mentions a literal apiKey last.
The shipped surfaces loaded the harness home's .env into the process
environment before cordis booted. credentials-local then saw every stored
key as an ambient launch override: describe reported source 'env' with
writable false, and set/unset rejected as shadowed — so a key the web page
or TUI stored was unrotatable and undeletable from the next run onward,
and the adapter kept using the value captured at launch.
The home's .env is now the credential provider's own store, read by that
provider alone and hot-reloaded by it. The genuine launch environment and
the invoking directory's .env (loaded by the bin) remain the read-only
ambient layer, so a plain composition without the provider still resolves
keys exactly as before.
Proven by a real restart in the loader composition: store a key through
the seam, dispose the tree, re-boot over the same harness home, and the
entry is still file-sourced and writable — rotating it lands on the very
next request.
The chat stats line took its token totals from the loaded conversation nodes,
so paging changed them and compaction erased the billing behind replaced
content. It also had no way to show context occupancy: the numerator and
capacity never reached the browser.
Both now come from token-meter session projections read through the standard
useProjection seat. Window nodes keep supplying turn and step counts plus LLM
and tool wall times, which are correctly window-scoped facts about what is on
screen; accounting no longer comes from there.
`tokenUsage` supplies billing and cache hit. `contextPressure` supplies
occupancy, pairing the newest provider-reported prompt size with the newest
capacity recorded by `request/context`. Deployments without token-meter drop
the token groups; a route whose adapter advertises no capacity drops the
occupancy group rather than rendering a placeholder.
Occupancy is deliberately approximate: the numerator and capacity are
independent last-wins fields, not one atomic request observation, so switching
models pairs a fresh capacity with the prior route's pressure until the next
request reports usage. It is a user-facing reference figure that nothing in the
harness makes decisions from, and it matches how the TUI status line has always
computed occupancy. The Agent Note and token-meter README state this as a
decision, including why the atomic alternative was implemented and rejected, so
it is not re-litigated as a defect.
Snapshot delta is one added `Context N% of 128K` segment across eight web
goldens; the preceding commit absorbed master's pre-existing golden drift.
Conflict resolution notes:
- StatsLine: master redesigned the row into pipe-separated groups with LLM and
tool wall times. Kept that design and swapped only the token accounting
source, so counts and durations stay window-scoped while billing and context
occupancy read the durable projections.
- Generated artifacts (cordis catalog, module graph, event producer/consumer,
i18n pairing hashes) and web snapshots took master's side; they are
regenerated and re-recorded after this merge.
- Web e2e goldens and details-panel/timeline assertions took master's side:
that evolution is unrelated to this branch.
- ui-conversation package.json: kept master's devDependency ordering, re-adding
only the token-meter entry this branch needs.
Replace the transient `session/model-request` mux frame with ordinary durable
session state. Occupancy now rides two last-wins projection fields instead of a
non-replayable frame that needed removal tombstones and cross-stream fencing.
The frame was the only non-replayable class on the mux stream. Because host and
mux are independent SSE streams with no cross-stream order, a request emitted
before a removal could arrive after `host/session-removed`, and a legitimate
request for a new lifecycle reusing the same id could be fenced by a late
removal. Fixing that needed a lifecycle generation on every frame; the frame
itself was the problem.
Removed: the `session/model-request` frame and schema, the `agent/model-request`
core event, the ApiProxy measurement point, the client-side telemetry map and
removal tombstone, and the synthetic `cancelled` open error used to signal
reconnect through the error channel.
Added: `request/context`, a log-only session event recording the
registration-bound capacity of the route a request resolved to, appended beside
`request/header` from the lookup that already prepared the call and skipped when
the route is unchanged. Capacity stays out of `EpochHeader` because it is
adapter metadata about a route, not an input the request was built from, so it
must not join request reconstruction or header equality.
The `contextPressure` projection pairs the newest provider-reported prompt size
with the newest recorded capacity. The two are deliberately not one atomic
request observation: switching models can pair a fresh capacity with the prior
route's pressure until the next request reports usage. The figure is a
user-facing reference, and this matches how the TUI status line has always
computed occupancy.
A plugin owning a standalone open/close bracket cannot tell a dead marker
from a live one: an unmatched `compact/start` reads identically whether the
previous writer died mid-compaction or a compaction is running now.
`Session.firstLiveSeq` already holds that answer exactly, but only in memory.
Append the log-only `session/inherited` event at that seq from the seeded
constructor — the single waist all six seeded-start paths pass through
(resume, configured startup on a persisted id, `sessions.fork()`, a subagent
fork child, `adopt()`'s live prefix, and a bare seeded `create`). Read it
through the new `isInheritedSeq(events, seq)`.
The constructor placement means persistence needs no changes: the marker is
already in `events` when a backend captures the creation seed, so it rides
the ordinary seed path with no load-time write. It also covers fork, where
the inherited bracket's owner may still be running — the case a
persistence-layer boundary could not reach.
Activity ordering excludes the boundary through `lastActivityTime()`, since
lazy resume makes browsing a pickup and the three call sites would otherwise
float every opened session to the top of a picker or list.
