Files
buzz/.env.example
Salman Mohammed 674c173eb7 feat(buzz-acp): give each channel thread its own agent session (#6732)
## What this does

In a channel, people often run several unrelated conversations at once
(separate threads). Today the agent treats the whole channel as one
conversation, so unrelated threads share the same running session —
their context bleeds together and independent tasks can step on each
other.

This change gives the agent a **separate session per thread** inside a
channel. Direct messages stay as one conversation (unchanged). The
channel is still the boundary for who is allowed in and what is visible
— only the agent's working context is now split by thread.

## How it is turned on

Off by default. Operators opt in with one setting:

- `BUZZ_ACP_SESSION_POLICY=channel` — default, current behavior
- `BUZZ_ACP_SESSION_POLICY=thread` — new per-thread behavior

Being behind a flag means we can enable it for a few agents, watch how
it behaves, and roll back instantly without a code change.

## Key design decisions

- **Decide the thread once, up front.** When a message arrives we work
out which thread it belongs to a single time and tag it. Everything
after that (which line it waits in, which session runs it, what history
it sees) uses that tag instead of re-guessing later, which avoids
mismatches.
- **Default stays identical to today.** Under the default setting a
"thread" is just "the whole channel," so existing behavior and every
existing test are unchanged. The new, riskier behavior is strictly
opt-in.
- **Give the agent only its thread's history.** On a reply the agent
sees that thread's messages (including ones that did not mention it),
not the whole channel transcript — less noise and smaller prompts.
- **Don't let one channel use more memory than before.** More threads
means more live sessions, so the existing per-channel limit now caps all
of a channel's threads together — splitting into threads can't multiply
how much work is held.

## Bugs found and fixed while iterating (from review)

- **Same thread, two sessions.** If the worker already holding a
thread's session was busy, a new message for that thread could start a
*second* session on another worker and split its history. Now it waits
for the right worker instead of forking.
- **Interrupting the wrong thread.** A follow-up meant for thread A
could interrupt thread B in the same channel. Interrupts now target the
exact thread.
- **Stuck thread after a crash.** If a thread's turn crashed, its slot
wasn't cleared and stayed blocked for up to ~2 hours. It now clears
right away and retries.
- **Lost the original request.** When a thread was interrupted and then
had to wait for a busy worker, only the follow-up was kept and the
original request was dropped. The full request is now preserved on
retry.
- **Same thread seen as two.** Two spellings of the same thread id
(upper/lower case) could be treated as different threads. Normalized so
they count as one.

## Not in this PR

- The desktop Settings toggle and rollout wiring for managed agents —
https://github.com/block/buzz/pull/6909
- One pre-existing retry edge case (present today without this flag,
unrelated to this change) — tracked separately so this PR stays focused.

## Testing

The full `buzz-acp` test suite passes (830+ unit and integration tests),
plus new focused tests for thread routing, session reuse, interrupt
targeting, crash recovery, and request preservation. Behavior with the
flag off is unchanged.

---------

Signed-off-by: Salman Mohammed <smohammed@squareup.com>
Signed-off-by: Leo <5faf251baee50ee6bcde338aef6acdd70bb3e60115664c2cd490d94a55dfc488@buzz.block.builderlab.xyz>
Co-authored-by: Leo <5faf251baee50ee6bcde338aef6acdd70bb3e60115664c2cd490d94a55dfc488@buzz.block.builderlab.xyz>
2026-08-31 18:17:28 -04:00

339 lines
17 KiB
Plaintext
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# =============================================================================
# Buzz Backend — Local Development Environment
# =============================================================================
# Copy this file to .env and adjust as needed:
# cp .env.example .env
#
# All defaults here work with `docker compose up` out of the box.
#
# Service ports (defaults):
# Postgres → localhost:5432
# Redis → localhost:6379
# Typesense → localhost:8108
# Adminer → localhost:8082 (DB browser UI)
#
# Note: If port 8082 conflicts, change the adminer port in docker-compose.yml
# =============================================================================
# -----------------------------------------------------------------------------
# Database (Postgres 17)
# -----------------------------------------------------------------------------
DATABASE_URL=postgres://buzz:buzz_dev@localhost:5432/buzz
# Optional read-replica URL; unset/blank keeps all reads on the writer.
# READ_DATABASE_URL=postgres://buzz:buzz_dev@localhost:5433/buzz
PGHOST=localhost
PGPORT=5432
PGUSER=buzz
PGPASSWORD=buzz_dev
PGDATABASE=buzz
# -----------------------------------------------------------------------------
# Redis 7
# -----------------------------------------------------------------------------
REDIS_URL=redis://localhost:6379
# Max connections in the relay's shared Redis pool (default 16).
# BUZZ_REDIS_POOL_SIZE=16
# Max connections in each of the relay's Postgres pools — writer and, when
# READ_DATABASE_URL is set, reader (default 50).
# BUZZ_DB_POOL_SIZE=50
# Writer-session Postgres timeouts for buzz-db-backed pools and the relay audit
# pool, all in milliseconds; 0 disables. The separately deployed push gateway
# owns its own database and session policy and does not consume these knobs.
# lock_timeout: fail a statement that waits this long on any lock instead of
# parking behind a wedged holder (default 5000).
# BUZZ_DB_LOCK_TIMEOUT_MS=5000
# idle_in_transaction_session_timeout: reap sessions idle inside an open
# transaction — bounds how long a wedged client can hold locks (default 60000).
# BUZZ_DB_IDLE_TXN_TIMEOUT_MS=60000
# statement_timeout: cap any single statement's runtime. Off by default —
# startup migrations/backfills legitimately run long statements. Warning: a
# pathologically low value (e.g. 1) also times out connection setup and can
# prevent any DB connection from establishing.
# BUZZ_DB_STATEMENT_TIMEOUT_MS=0
# -----------------------------------------------------------------------------
# Typesense (search)
# -----------------------------------------------------------------------------
TYPESENSE_API_KEY=buzz_dev_key
TYPESENSE_URL=http://localhost:8108
# -----------------------------------------------------------------------------
# Relay (WebSocket server)
# -----------------------------------------------------------------------------
# Bind address for the relay (host:port)
BUZZ_BIND_ADDR=0.0.0.0:3000
# Public WebSocket URL — used in NIP-42 auth challenges
RELAY_URL=ws://localhost:3000
# Stable relay signing key (required). `just bootstrap` generates a random key in
# the gitignored .env file. Preserve that value across restarts and backups.
# BUZZ_RELAY_PRIVATE_KEY=<32-byte hex private key>
# Optional: path to the web UI dist directory. When set, the relay serves
# the web frontend at / for browser requests. Leave unset for local dev
# (use `just web` for Vite HMR instead).
# BUZZ_WEB_DIR=./web/dist
# NIP-PL mobile push is an explicit deployment opt-in. A gateway URL alone
# never enables it. When enabled and the URL is absent, the canonical
# https://push.buzz.xyz/v1/deliveries/apns endpoint is used.
BUZZ_PUSH_ENABLED=false
# BUZZ_PUSH_GATEWAY_DELIVERY_URL=https://push.buzz.xyz/v1/deliveries/apns
# -----------------------------------------------------------------------------
# Admin Dashboard (private moderation surface)
# -----------------------------------------------------------------------------
# Host name that serves the moderation dashboard and its /api/admin/v1
# endpoints. Leave unset to keep the admin surface absent.
# BUZZ_ADMIN_HOST=admin.localhost:3000
#
# Authentication mode. Accepted values: nip98 (default), disabled.
# Any other value is a startup error. Token authentication was removed:
# BUZZ_ADMIN_TOKEN is ignored with a startup warning — remove it from the environment.
# BUZZ_ADMIN_AUTH=nip98
#
# Option A — BUZZ_ADMIN_AUTH=nip98 (Nostr pubkey-based auth, default):
# NIP-98 HTTP Auth. Each request must carry an Authorization: Nostr header
# with a signed kind-27235 event. Authorized principals are resolved from:
# 1. RELAY_OPERATOR_PUBKEYS — comma-separated 64-char hex pubkeys (config Operators).
# 2. RELAY_OWNER_PUBKEY — implicit Operator fallback when RELAY_OPERATOR_PUBKEYS is unset.
# 3. relay_operators table — DB-managed Operator/Moderator roster.
# The dashboard requires a NIP-07 browser extension.
# Setting RELAY_OPERATOR_PUBKEYS for the admin console does NOT require
# RELAY_OPERATOR_API_ORIGIN; that origin is only for community provisioning
# (see below). When BUZZ_ADMIN_HOST is set, the relay advertises the admin
# origin in its NIP-11 document (`admin_api` field) so clients can auto-discover
# the console without manual URL entry.
# RELAY_OPERATOR_PUBKEYS=<64-char hex pubkey>[,<64-char hex pubkey>...]
#
# Option B — BUZZ_ADMIN_AUTH=disabled (network-layer auth only):
# Set only when the admin API is already protected at the network layer
# (VPN, private ingress). The relay logs a WARN on every startup.
# `just admin` defaults to this mode for local review.
#
# Directory holding the built dashboard assets (`pnpm -C admin-web build`).
# BUZZ_ADMIN_WEB_DIR=./admin-web/dist
#
# Canonical origin (http(s)://host[:port], no path) that community-provisioning
# NIP-98 requests are verified against. Required only to USE the provisioning
# endpoints (POST /operator/communities) — not for the admin console. When
# RELAY_OPERATOR_PUBKEYS is set but this is unset, the relay boots with a WARN
# and provisioning requests fail closed until it is set.
# RELAY_OPERATOR_API_ORIGIN=http://127.0.0.1:3000
# Optional relay-owned KLIPY key. When set, NIP-11 advertises GIF search and
# authenticated desktop clients use this relay as the metadata/search proxy.
# Keep the real value in your deployment's secret manager; never commit it.
# BUZZ_KLIPY_API_KEY=
# Shared Redis-backed admission limits. Defaults shown below; each value must
# be a positive integer.
# BUZZ_RATE_LIMIT_HUMAN_MESSAGES_PER_MIN=60
# BUZZ_RATE_LIMIT_GIF_SEARCHES_PER_MIN=30
# BUZZ_RATE_LIMIT_HUMAN_API_CALLS_PER_MIN=300
# BUZZ_RATE_LIMIT_HUMAN_WS_EVENTS_PER_SEC=10
# BUZZ_RATE_LIMIT_AGENT_STANDARD_MESSAGES_PER_MIN=120
# BUZZ_RATE_LIMIT_AGENT_STANDARD_API_CALLS_PER_MIN=600
# BUZZ_RATE_LIMIT_AGENT_ELEVATED_MESSAGES_PER_MIN=300
# BUZZ_RATE_LIMIT_AGENT_PLATFORM_MESSAGES_PER_MIN=600
# -----------------------------------------------------------------------------
# Git (NIP-34 bare repositories)
# -----------------------------------------------------------------------------
# Root directory for ephemeral Git workspaces and the disposable pack cache.
# Default: ./repos (relative to CWD).
# BUZZ_GIT_REPO_PATH=./repos
# BUZZ_GIT_MAX_PACK_BYTES=524288000
# BUZZ_GIT_MAX_REPO_BYTES=1048576000
# Process-local immutable pack/index cache. Zero disables retention.
# BUZZ_GIT_PACK_CACHE_PATH=./repos/.pack-cache
# BUZZ_GIT_PACK_CACHE_MAX_BYTES=5368709120
# BUZZ_GIT_PACK_CACHE_MAX_CONCURRENT_POPULATIONS=2
# -----------------------------------------------------------------------------
# S3-Compatible Object Storage (media + Git/CAS)
# -----------------------------------------------------------------------------
# The local MinIO container is reachable from host processes at localhost:9000.
# Path style keeps the bucket in the URL path and is required by this local DNS
# setup. Use `virtual` only when the provider requires bucket-as-subdomain URLs.
BUZZ_S3_ENDPOINT=http://localhost:9000
BUZZ_S3_ACCESS_KEY=buzz_dev
BUZZ_S3_SECRET_KEY=buzz_dev_secret
BUZZ_S3_BUCKET=buzz-media
BUZZ_S3_REGION=us-east-1
BUZZ_S3_ADDRESSING_STYLE=path
# -----------------------------------------------------------------------------
# Media Upload Admission
# -----------------------------------------------------------------------------
# Bound image/video parser and storage work admitted by one relay process.
# BUZZ_MEDIA_MAX_CONCURRENT_UPLOADS=8
# BUZZ_MEDIA_MAX_CONCURRENT_UPLOADS_PER_PUBKEY=2
# BUZZ_MEDIA_UPLOADS_PER_MINUTE=30
# GET/HEAD /media/* always require Blossom t=get auth and relay membership.
# BUZZ_REQUIRE_MEDIA_GET_AUTH and BUZZ_REQUIRE_MEDIA_READ_AUTH are no longer
# read; setting either (including to false) changes nothing and the relay warns
# about it at startup.
# -----------------------------------------------------------------------------
# Ephemeral Channels (TTL testing)
# -----------------------------------------------------------------------------
# Override the TTL for all ephemeral channels (in seconds). When set, any
# channel created with a TTL tag will use this value instead of the
# client-provided one. Unset to use the client-provided TTL.
# BUZZ_EPHEMERAL_TTL_OVERRIDE=60
# How often the reaper checks for expired ephemeral channels (default: 60s).
# BUZZ_REAPER_INTERVAL_SECS=5
# -----------------------------------------------------------------------------
# Logging / Tracing
# -----------------------------------------------------------------------------
RUST_LOG=buzz_relay=debug,buzz_datastore=info,buzz_db=debug,buzz_auth=debug,buzz_pubsub=debug,tower_http=debug
# Optional OpenTelemetry-only target filter. This is deliberately independent
# from RUST_LOG so log verbosity changes cannot break trace parentage.
# BUZZ_OTEL_FILTER=buzz_relay=info,buzz_datastore=info
# OTLP tracing endpoint (optional — leave unset to disable)
# OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
# -----------------------------------------------------------------------------
# ACP (Agent Communication Protocol — buzz-acp harness)
# -----------------------------------------------------------------------------
# The ACP harness bridges Buzz events to AI agents. Each env var below maps
# to a CLI flag of the same name (lowercase, hyphens → underscores). All values
# are optional unless noted; defaults are shown in comments.
#
# Quick start:
# BUZZ_PRIVATE_KEY=<hex> BUZZ_RELAY_URL=ws://localhost:3000 buzz-acp
# ── Identity & auth ──────────────────────────────────────────────────────────
# Nostr private key (hex or bech32). REQUIRED — identifies the agent on the relay.
# BUZZ_PRIVATE_KEY=<32-byte hex or nsec1… private key>
# Relay WebSocket URL the harness connects to.
# Note: the relay itself uses RELAY_URL (above); this is the ACP harness's
# connection target — they happen to point at the same place in local dev.
# BUZZ_RELAY_URL=ws://localhost:3000
# ── Agent subprocess ─────────────────────────────────────────────────────────
# Binary to spawn as the AI agent (e.g. "goose", "codex-acp", "claude-code").
# BUZZ_ACP_AGENT_COMMAND=goose
# Comma-separated arguments passed to the agent binary.
# Goose default: "acp". Codex/Claude default: "" (empty).
# BUZZ_ACP_AGENT_ARGS=acp
# Binary for an optional MCP server sidecar (e.g. buzz-dev-mcp for buzz-agent).
# BUZZ_ACP_MCP_COMMAND=
# Number of parallel agent subprocesses (1–32).
# BUZZ_ACP_AGENTS=1
# Desired LLM model ID. Applied to every new ACP session.
# Use `buzz-acp models` to discover available model IDs.
# BUZZ_ACP_MODEL=
# Optional Databricks model-picker visibility filter. Discovery-only; this does
# not grant inference access. Comma-separated full-string * / ? patterns are
# OR-matched against raw workspace endpoint and Unity Catalog model-service IDs.
# Unset or blank shows every catalog entry. A nonblank value with no usable patterns is invalid.
# DATABRICKS_MODEL_FILTER=databricks-*,data_tools.goose.*
# ── Timeouts & sessions ──────────────────────────────────────────────────────
# Max seconds per agent turn before timeout (default 320 = ~5 min).
# BUZZ_ACP_TURN_TIMEOUT=320
# Max turns per session before proactive rotation. 0 = disabled (rotate only
# on MaxTokens / MaxTurnRequests). Recommended: 50 for long-running agents.
# BUZZ_ACP_MAX_TURNS_PER_SESSION=0
# ── Prompts ──────────────────────────────────────────────────────────────────
# System prompt injected into every agent session (inline text).
# BUZZ_ACP_SYSTEM_PROMPT=
# Path to a file containing the system prompt (mutually exclusive with above).
# BUZZ_ACP_SYSTEM_PROMPT_FILE=
# Message sent to the agent immediately after session creation.
# BUZZ_ACP_INITIAL_MESSAGE=
# ── Heartbeat ────────────────────────────────────────────────────────────────
# Seconds between heartbeat prompts. 0 = disabled. Must be 0 or ≥10.
# Recommended: 60 for long-running agents to prevent idle session timeouts.
# BUZZ_ACP_HEARTBEAT_INTERVAL=0
# Heartbeat prompt text (inline). Mutually exclusive with file variant.
# BUZZ_ACP_HEARTBEAT_PROMPT=
# Path to a file containing the heartbeat prompt.
# BUZZ_ACP_HEARTBEAT_PROMPT_FILE=
# ── Desktop development ──────────────────────────────────────────────────────
# DEV-only: replay first-run onboarding and the Welcome Team kickoff on each
# app launch while keeping the current identity and relay data.
# VITE_BUZZ_FORCE_FRESH_ONBOARDING=true
# Protected internal builds only: selects the module graph that contains the
# default-off Bestie experiment. Official OSS builds must leave this unset.
# VITE_BUZZ_BESTIE=1
# ── Subscription & filtering ─────────────────────────────────────────────────
# Subscribe mode: "mentions" (default), "all", or "config" (rule-based).
# BUZZ_ACP_SUBSCRIBE=mentions
# Comma-separated event kind numbers to subscribe to (overrides mode defaults).
# BUZZ_ACP_KINDS=
# Comma-separated channel UUIDs to limit subscription scope.
# BUZZ_ACP_CHANNELS=
# Set to true to disable the @-mention filter in mentions mode.
# BUZZ_ACP_NO_MENTION_FILTER=false
# Path to TOML config file for rule-based subscriptions (config mode).
# BUZZ_ACP_CONFIG=./buzz-acp.toml
# ── Dedup & self-ignore ──────────────────────────────────────────────────────
# How to handle duplicate events: "queue" (default) or "drop".
# BUZZ_ACP_DEDUP=queue
# Set to true to process the agent's own messages (default: ignore self).
# BUZZ_ACP_NO_IGNORE_SELF=false
# ── Session scoping ──────────────────────────────────────────────────────────
# How ACP provider sessions are scoped in channels: "channel" (default) or
# "thread". "channel" keeps one provider session per channel (legacy). "thread"
# gives each canonical channel thread its own isolated provider session; direct
# messages stay conversation-scoped either way. Ships as "channel" so thread
# scoping can be canaried and rolled back without code changes.
# BUZZ_ACP_SESSION_POLICY=channel
# ── Context ──────────────────────────────────────────────────────────────────
# Max context messages fetched for thread replies and DMs (0–100). 0 = disabled.
# BUZZ_ACP_CONTEXT_MESSAGE_LIMIT=12
# ── Presence & typing ────────────────────────────────────────────────────────
# Set to true to disable automatic online/offline presence status.
# BUZZ_ACP_NO_PRESENCE=false
# Set to true to disable typing indicators while the agent is processing.
# BUZZ_ACP_NO_TYPING=false
# ── Advanced tuning ──────────────────────────────────────────────────────────
# Event channel buffer capacity (WebSocket → harness). Increase for
# high-throughput agents. Minimum 1.
# BUZZ_ACP_EVENT_BUFFER=256
# ── Legacy aliases ───────────────────────────────────────────────────────────
# These are accepted for backward compatibility but the canonical names above
# are preferred:
# BUZZ_ACP_PRIVATE_KEY → BUZZ_PRIVATE_KEY
# Optional relay join policy. Markdown is served by the relay so every join
# surface can present the same documents. Each document and the independent age
# attestation are optional; configuring any one enables policy acceptance.
# BUZZ_TERMS_OF_SERVICE_MARKDOWN="# Terms of Service\n\nFull terms here."
# BUZZ_PRIVACY_POLICY_MARKDOWN="# Privacy Policy\n\nFull policy here."
# BUZZ_AGE_ATTESTATION_REQUIRED=true