The support chat showed 'Tänker' and then went quiet with no answer and
no error. Root cause: the 2026-08-21 RIP-3 cutover moved general.help to
the single-call POST /api/agent/ask, whose 1500-token default cap made
stop_reason max_tokens routine on tool-loop turns. The empty answer then
passed unlogged through the service, the route answered 200 with an
empty string, and the console appended an invisible empty bubble.
Fixes, single-call path:
- ask-service: default maxTokens 1500 -> 5400 (the streaming chat's
reply ceiling); an empty final answer now logs model + usage and
throws the typed EmptyModelAnswerError instead of passing through.
- /api/agent/ask: maxDuration 300, logger, empty answer maps to 502
with 'Assistenten gav inget svar. Försök igen.'; an empty assistant
turn is never persisted (the question stays, so retry works).
- AskConsole: a 200 with an empty answer shows the error box instead of
appending an invisible bubble.
- anthropic-family: serialized tool results are bounded at 40000 chars
(mirrors run-turn) so one big read cannot eat the output budget; the
step-exhausted fallback keeps tools declared with tool_choice none,
because replaying tool_use/tool_result without tools is an API 400.
Fixes, streaming path (same silent class):
- /api/agent/invoke: maxDuration 300 so deep thinking turns are not
killed mid-stream at the platform default cap.
- run-turn: stop_reason max_tokens with no visible text emits an error
event, not a bare turn_complete.
- AgentChat: an NDJSON stream that ends without turn_complete or error
(and was not aborted) shows 'Anslutningen bröts innan svaret blev
klart. Försök igen.'
The lib/ai request-shape tests were updated deliberately for the
fallback change; general.help stays on the single-call runtime
(founder decision, not reverted).
Claude-Session: https://claude.ai/code/session_01SyDuePXxUFowaPBKpAv8SF
Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The /chat assistant (audit Option A / rip) shipped in #1759 reading only the
company name + entity type, so it answered "jag har ingen bokföringsdata" to
every figures question ("vad är min största utgiftspost?"). It now behaves like
an MCP client: it answers over a bounded, READ-only tool loop across the same
MCP read tools the old streaming assistant had, plus an always-on company
snapshot as the backstop.
Provider-agnostic by construction, so it still runs on a local model:
- lib/ai generateText gains optional `tools` + `maxSteps`. The OpenAI-compatible
service forwards them to the Vercel AI SDK (stopWhen: stepCountIs), which runs
the loop; the Anthropic-family service hand-rolls a small loop against
messages.create. Kept on the raw Anthropic SDK: no new deps, and the no-tools
path is byte-identical, so hosted extraction/composer/etc. are unchanged.
- lib/agent/ask/ledger-tools.ts: the read slice of general.help's whitelist
(income statement, VAT, ledgers, query_journal, reskontror, lists…) from
agentToolRegistry, dispatched with the agent_chat actor run-turn uses. Write/
staging + memory-write tools are excluded; readOnlyHint/destructiveHint are
re-checked. Empty in a core-only build → snapshot-only, graceful.
- lib/agent/ask/snapshot.ts: a compact company_settings + deadlines block so a
model that can't/won't call tools still answers status questions. Never carries
figures (those come from the live tools).
- ask-service attaches tools + snapshot when a userId is present and uses a
tool-aware system prompt; the route calls ensureInitialized() so the registry
is populated and threads userId/conversationId through.
Works on Bedrock and on any local model with function-calling (Qwen). Tests:
the anthropic hand-rolled loop (tool call → result → answer, is_error handling,
step-budget forced answer), openai tool forwarding, the read-only adapter
filter, the snapshot format, and the ask-service wiring. 457 agent+ai tests
green, lint/guards clean.
Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
RIP-3 cutover. The free-form /chat panel (general.help) now answers through a
page-scoped single-call console (AskConsole → POST /api/agent/ask) instead of
the streaming Anthropic runtime, so the in-app assistant runs on ANY configured
backend, including a local OpenAI-compatible model (Qwen behind
llama.cpp/Ollama/vLLM). No tool loop, no NDJSON stream, no Anthropic wire format.
Threads still persist: the ask route gains an opt-in persist branch that writes
both turns to agent_conversations/agent_messages as canonical Anthropic text
blocks, so the /chat sidebar and "resume a thread" keep working across old
streaming threads and new single-call ones. Page-scoped one-off asks (a report
page) omit persist and stay stateless.
Scope: only general.help is wired to the console. The tool-loop intents
(transaction.categorization, invoice.draft, supplier_invoice.review) and the
docked AgentSheet still use AgentChat + run-turn.ts because they stage
operations and need the tool loop, so run-turn.ts is intentionally NOT deleted
here (the plan gates its deletion on "once nothing calls them"; RIP-4 migrates
the rest).
- lib/agent/ask/persist.ts: resolveChatConversation (create/resume, ownership),
persistUserTurn, persistAssistantTurn (append + roll last_message_* forward)
- app/api/agent/ask/route.ts: persist branch (resolve → user turn → answer →
assistant turn), returns conversation_id; 404 on a foreign conversation
- components/agent/AskConsole.tsx: the console UI (approved sign-off design):
user bubble + bare-prose answer, thinking indicator, empty/503/paywall states
- ChatConversationView / ChatNewStarter: branch general.help → AskConsole,
every other intent keeps AgentChat unchanged
Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
WS1 rip track, RIP-2 (audit Option A), stacked on the AI provider
abstraction (#1740). The first replacement for the streaming Anthropic chat
runtime: a page-scoped, single-call assistant answer.
- lib/agent/ask/ask-service.ts: answerAssistantQuestion() uses
getAiService().generateText, so it runs on whatever backend is configured:
AWS Bedrock, the direct Anthropic API, OR any OpenAI-compatible endpoint,
including a local model (Qwen behind llama.cpp/Ollama/vLLM). No tool loop,
no Anthropic wire format, nothing to translate per provider. The caller (a
page) supplies the context; the service reads only the company's own
profile for grounding, and the system prompt forbids inventing figures.
- POST /api/agent/ask: same auth/rate-limit/sandbox/paywall gates as
/invoke, but gated on getAiStatus().configured (not assistantAvailable),
because ANY provider works here. That is the difference that lets the
assistant answer on a local model where the streaming /invoke returns 503.
This is the non-UI foundation of the rip: the thin /chat console and the
page-scoped actions (RIP-3, UI, gated on visual sign-off) will consume this
endpoint; run-turn.ts's streaming path and the intents' getAnthropic() usage
are removed once nothing calls them.
Verified: 10 unit tests (service prompt shape + tier + context-as-data +
truncation; route 401/429/400/403 paywall/200-on-openai-compatible/503
unconfigured) + a live smoke against a local OpenAI-compatible mock (resolved
provider openai-compatible, POSTed model qwen3.8 with a placeholder key,
returned an answer). 421 agent/ai tests green; tsc, guards, lint clean.
Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
