dguiducci
baf68878e4
Nightly Build / build (push) Successful in 7m35s
The shipped default combined a sliding history window with no compaction, which is the worse of the two available trades in both directions it is measured on. `max_history_messages: 30` is a sliding tail window (`projection::window` — `drain(..len - max)`). Past 30 messages it drops from the head on *every* turn, so the prompt prefix changes on every single request and every provider that caches one (Anthropic breakpoints, OpenAI automatic prefix caching) misses every time. It also drops those messages with no summary standing in for them: silent amnesia, not just a cold cache. Compaction rewrites the prefix once per compaction and leaves a summary behind — yet it was the half that was commented out, while the window's own doc-comment already said the two were exclusive. Both are now `Option` and both ship unset, so nothing shrinks a conversation unless a human types `/compact`. Which surfaced the real bug: `/compact` did not work either. The compactor was `Option<Arc<ContextCompactor>>` keyed on the config section existing, so commenting out `compaction:` disabled the manual command too — `force_compact` returned `Ok(false)` and the chat answered "compaction disabled". Manual compaction is a command a user types; it cannot depend on an admin having filled in a token threshold. The compactor is now built unconditionally and `threshold_tokens: Option<u32>` arms only the automatic pass; `try_compact` early-returns without it, `force_compact` deliberately never consults it. The projection accordingly yields to the *automatic* pass rather than to the compactor's existence (`LoopConfig.auto_compaction_enabled`), so a configured message cap is not silently voided by `/compact` merely being available. `CompactionConfig::Default` is hand-written for the same reason `RoleAttrs`'s is: a derived one gives `keep_recent: 0`, which would compact away every recent message on any box omitting the section — now the default. Also fixes two documentation bugs in the same file: `event_triage` was documented nested under `llm:`, where it parses fine and is then silently ignored (it is a top-level field), and `datetime` was documented twice with conflicting examples. A new test asserts the shipped default actually deserializes and that both guards are off — a field the default omits must be genuinely optional, or a brand-new install fails to boot. Automatic compaction returns later, triggered off the resolved model's own context window instead of a hand-tuned token count that cannot know which model is answering.
167 lines
10 KiB
YAML
167 lines
10 KiB
YAML
server:
|
|
host: 127.0.0.1
|
|
port: 9000
|
|
|
|
# ── Global timezone ─────────────────────────────────────────────────────────────
|
|
# IANA timezone name applied globally to:
|
|
# - cron expression evaluation (next_run_at computation)
|
|
# - the date/time string injected into the LLM context each turn
|
|
# When omitted, the server's local system timezone is used.
|
|
# Examples: Europe/Rome, Europe/London, America/New_York, Asia/Tokyo
|
|
#
|
|
# timezone: Europe/Rome
|
|
# ────────────────────────────────────────────────────────────────────────────────
|
|
|
|
web:
|
|
static_dir: ./web
|
|
|
|
# ── Connector marketplace ──────────────────────────────────────────────────────
|
|
# The feed of vetted connectors the admin browses under Connectors → Marketplace.
|
|
# It is *consultative*: the feed proposes, the admin installs into the local
|
|
# catalog, and only then can a connector be enabled globally or activated by a
|
|
# user. The trust anchor stays on this box.
|
|
#
|
|
# Point it at a self-hosted mirror or a local copy to run fully offline — the feed
|
|
# is plain static files (`connectors.json` + `<folder>/connector.json`).
|
|
marketplace:
|
|
url: https://connectors.skaldagent.net
|
|
|
|
# The database lives at ./database/system.db — fixed, not configurable.
|
|
|
|
|
|
# ── LLM clients ────────────────────────────────────────────────────────────────
|
|
# LLM clients (providers, models, API keys, strength) are configured
|
|
# via the web app and stored in the database — not in this file.
|
|
# ───────────────────────────────────────────────────────────────────────────────
|
|
llm:
|
|
# ── History window (DISABLED by default) ────────────────────────────────────
|
|
# Hard cap on the number of history messages sent to the LLM, applied as a
|
|
# sliding tail window: past the cap, the oldest messages are dropped.
|
|
#
|
|
# Off by default, for two reasons:
|
|
# 1. Cache. Once history exceeds the cap, every turn shifts the window's
|
|
# start, so the prompt prefix changes on every single request and the
|
|
# provider's prompt cache (Anthropic breakpoints, OpenAI automatic prefix
|
|
# caching) misses every time. Append-only history keeps the prefix stable
|
|
# for the whole conversation.
|
|
# 2. Memory. The window drops messages with no summary standing in for them,
|
|
# so the assistant silently forgets. `/compact` replaces them with a
|
|
# summary instead.
|
|
#
|
|
# With this off and automatic compaction off (the shipped default), the context
|
|
# grows until the model's own limit — use `/compact` to summarise it.
|
|
# Ignored when automatic compaction is enabled (see `compaction` below).
|
|
#
|
|
# max_history_messages: 30
|
|
# ───────────────────────────────────────────────────────────────────────────
|
|
max_tool_rounds: 100
|
|
# Max synchronous sub-agents dispatched concurrently when the LLM emits a
|
|
# homogeneous batch (≥2) of sub-agent calls (execute_task mode=sync /
|
|
# execute_subtask) in a single response. Bounds fan-out to avoid provider
|
|
# rate-limit storms. Omit for the default (4); set to 1 to force sequential.
|
|
max_parallel_subagents: 4
|
|
datetime:
|
|
enabled: true
|
|
round_minutes: 60 # Help with KV cache (instead of 10:54, it will pass 10:50 to the LLM)
|
|
|
|
# ── Tool result size limit ──────────────────────────────────────────────────
|
|
# When set, tool results from *previous* turns that exceed this character
|
|
# count are replaced (at context-build time) with a short placeholder:
|
|
# "[Tool response hidden: N chars. Call the tool again if you need it.]"
|
|
# The original result is always preserved in the database and shown in the
|
|
# frontend; only what the LLM receives in subsequent turns is affected.
|
|
# The current turn always sees full results regardless of this limit.
|
|
# Omit or comment out to disable (no limit).
|
|
max_tool_result_chars: 10000
|
|
# ───────────────────────────────────────────────────────────────────────────
|
|
|
|
# ── Context compaction (AUTOMATIC pass disabled by default) ─────────────────
|
|
# Compaction summarises old history into a single block, persisted to the DB
|
|
# and injected at the start of subsequent turns in place of the messages it
|
|
# covers, keeping the last `keep_recent` raw messages for immediate context.
|
|
#
|
|
# The `/compact` command works ALWAYS and needs nothing here — this whole
|
|
# section is optional and only tunes it.
|
|
#
|
|
# `threshold_tokens` is what arms the AUTOMATIC pass: set it, and history is
|
|
# compacted on its own once the previous turn exceeded that many input tokens.
|
|
# It is COMMENTED OUT by default: every compaction rewrites the prompt prefix
|
|
# and so costs a prompt-cache miss, and doing it unprompted trades away context
|
|
# the user may still need. Compact manually with `/compact` for now.
|
|
# (Future: an automatic pass triggered by the model's own context window rather
|
|
# than by a hand-tuned token count.)
|
|
#
|
|
# `strength` controls which LLM is picked for summary generation via the AUTO
|
|
# selector (same strength levels used for agent assignment). Compaction is a
|
|
# simple writing task — `low` or `average` is usually sufficient.
|
|
# Omit `strength` to use whatever AUTO picks.
|
|
# NOTE: the Settings page has an instance-wide "Compaction model" picker
|
|
# (registry config key `compaction_model`) — when set, it wins over this
|
|
# `strength` fallback and needs no restart.
|
|
#
|
|
# When the LLM provider does not report token usage (e.g. some LM Studio
|
|
# setups), a rough estimate (total chars / 4) is used as a fallback.
|
|
#
|
|
# compaction:
|
|
# threshold_tokens: 30000 # arms the automatic pass, above this many input tokens
|
|
# keep_recent: 6 # raw messages kept outside the summary
|
|
# strength: low # LLM strength for summary generation
|
|
# ───────────────────────────────────────────────────────────────────────────
|
|
|
|
# ── Date/time injection ─────────────────────────────────────────────────────
|
|
# Configured above as `datetime`. Rounding keeps the injected timestamp stable
|
|
# for up to N minutes, so the dynamic tail can be KV-cached across requests
|
|
# instead of changing every second.
|
|
# enabled: true # set to false to disable injection entirely
|
|
# round_minutes: 10 # round down to nearest N minutes (e.g. 10:56 → 10:50)
|
|
# ───────────────────────────────────────────────────────────────────────────
|
|
|
|
# ── LLM request/response log ────────────────────────────────────────────────
|
|
# Logs every chat_with_tools call to the `llm_requests` table.
|
|
# Captures the full HTTP request body, response body, headers (api-key redacted),
|
|
# token counts, and round-trip duration.
|
|
#
|
|
# WARNING: disabling `enabled` also disables the home-page LLM statistics.
|
|
#
|
|
# What to save (all default to true):
|
|
# request_payload_save — full request JSON (can be hundreds of KB per call)
|
|
# response_payload_save — full response JSON
|
|
# request_header_save — HTTP request headers (api-key always redacted)
|
|
# response_header_save — HTTP response headers
|
|
#
|
|
# Cleanup policy (all optional — omit to keep data forever):
|
|
# cleanup_request_payload_after — set request_json = '' after N days
|
|
# cleanup_response_payload_after — set response_json = NULL after N days
|
|
# cleanup_headers_after — null out both header columns after N days
|
|
# cleanup_rows_after — physically delete rows after N days
|
|
#
|
|
# The cleaner runs 1 minute after startup, then every 12 hours.
|
|
#
|
|
requests_log:
|
|
enabled: true
|
|
request_payload_save: false
|
|
response_payload_save: true
|
|
request_header_save: true
|
|
response_header_save: true
|
|
cleanup_request_payload_after: 7
|
|
cleanup_response_payload_after: 14
|
|
cleanup_headers_after: 30
|
|
cleanup_rows_after: 90
|
|
# ───────────────────────────────────────────────────────────────────────────
|
|
|
|
# ── Event triage (background event processor) ──────────────────────────────────
|
|
# Runs periodically to process pending MCP events (email, calendar, WhatsApp) and
|
|
# decide whether to surface a notification to the user.
|
|
#
|
|
# NOTE: top-level, NOT under `llm:` — nesting it there parses fine and is then
|
|
# silently ignored.
|
|
#
|
|
# interval_secs — how often it runs (default: 900 = 15 minutes)
|
|
# batch_size — max events processed per pass (default: 50)
|
|
#
|
|
# event_triage:
|
|
# interval_secs: 900
|
|
# batch_size: 50
|
|
# ───────────────────────────────────────────────────────────────────────────────
|
|
|