feat(telemetry): local-first telemetry & observability

Add a built-in telemetry system that records what the agent does — workflows,
model calls, tool calls, errors — to the local machine, powers `/insights`, and
can export to an operator-chosen destination. Default-on locally; nothing leaves
the machine unless the user exports it or opts into the aggregate plane.

Three planes with a hard wall between them:
  - local: full-fidelity observability (real model/provider/tool names), on by
    default, never leaves the machine.
  - aggregate: opt-in metadata, default off. No uploader ships — consent is
    recorded via telemetry.consent_state, and `preview` shows what would be
    produced, computed locally.
  - trajectories: full message content, opt-in, exported only to the operator's
    own destination.

Mechanism:
  - Bundled `telemetry` plugin registers observational lifecycle hooks
    (on_session_start / post_api_request / post_tool_call / on_session_finalize).
    No core call sites are edited; hooks already carry the data.
  - Fire-and-forget emitter: emit() returns in microseconds, never blocks or
    raises into a model/tool call. A daemon thread writes events to an
    append-only JSONL log and the tel_* tables in state.db (its own sqlite
    connection, separate from SessionDB).
  - tel_runs / tel_model_calls / tel_tool_calls live in the declarative
    SCHEMA_SQL and are reconciled automatically; SCHEMA_VERSION 16 -> 17.
  - metrics derives rollups for /usage and /insights; rollup builds per-run
    summaries for `hermes telemetry preview`.

Consent is config, not a parallel command surface. The config file is the root
of trust: set telemetry.consent_state with `hermes config set`, or pin any
telemetry.* key (including allow_aggregate) via managed scope, which overrides
the user's value per key. `hermes telemetry` exposes only what config cannot:
status (report), preview (query), and export.

Export:
  - exporter_bulk writes telemetry (and, when the trajectories plane is enabled,
    session content) to ndjson/json.
  - otlp_exporter streams spans to a configured OpenTelemetry Collector over
    OTLP/HTTP. The SDK is an optional extra (hermes-agent[otlp]), lazily
    installed via tools.lazy_deps on first use.
  - Secrets are always redacted on every export path
    (redact_sensitive_text(force=True)); content export is gated by the
    trajectories plane, and PII scrubbing follows telemetry.content_redaction.
    OTLP auth headers reference environment variable names, never inline values.

No outbound emission to Nous. The aggregate uploader is intentionally not built.
This commit is contained in:
emozilla 2026-06-24 02:14:02 -04:00
parent 64131bf975
commit ccfa079252
37 changed files with 3920 additions and 4 deletions

View file

@ -1159,6 +1159,49 @@ display:
# # Routing/delivery still uses the original values internally.
# redact_pii: false
# =============================================================================
# Telemetry & Observability
# =============================================================================
# Three planes with a hard wall between them:
# local — full-fidelity observability you own. Default ON.
# aggregate — opt-in metadata to Nous (no uploader ships yet). Default OFF.
# trajectories — content trajectories for training. Separate consent (later).
#
# The local plane records real values (actual models, providers, tool names) —
# your own data, on your machine. The aggregate plane is opt-in and has no
# uploader today; if one ships it would summarize at that egress boundary.
#
# Enterprise locking: any telemetry.* key can be pinned by an administrator via
# the managed-scope layer (/etc/hermes/config.yaml), which wins over the user's
# value. To hard-forbid egress on a locked-down deployment, pin
# `telemetry.allow_aggregate: false` there.
# telemetry:
# # Local plane: event log + SQLite index in state.db. Never leaves your
# # machine unless you export it or opt into the aggregate plane.
# local: true
# # Hard gate for the aggregate plane. When false, the aggregate plane is off
# # regardless of consent_state. Pin false via managed scope to forbid egress.
# allow_aggregate: true
# # Aggregate-plane consent (the opt-in). No uploader ships yet.
# # unknown (no choice — never uploads) | local (declined) | aggregate (opted in)
# consent_state: unknown
# # Stable install id (aggregate plane only). Empty = mint on first use;
# # clear it to rotate.
# install_id: ""
# # Local event-log retention before rotation (days).
# retention_days: 90
# # Keep secret redaction on even at full local capture.
# redact_secrets: true
# # Content redaction for exports / support bundles: none | pii.
# content_redaction: none
# # Exporters. The OTLP exporter sends spans to a configured Collector endpoint;
# # header values reference environment variable names, not inline secrets.
# export:
# otlp:
# enabled: false
# endpoint: null
# headers_env: {} # e.g. {Authorization: MY_OTLP_TOKEN_ENVVAR}
# =============================================================================
# Shell-script hooks
# =============================================================================