2025-08-01 21:19:30 -04:00
|
|
|
# LLM Proxy Configuration Example
|
|
|
|
|
# This file demonstrates all available configuration options
|
|
|
|
|
# Copy this file to config.yaml and customize as needed
|
|
|
|
|
|
|
|
|
|
# Server configuration
|
|
|
|
|
server:
|
2026-03-19 19:00:24 -06:00
|
|
|
# Bind host for the proxy server.
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# Example local-only value. The current built-in default is 0.0.0.0, but
|
|
|
|
|
# startup validation rejects public binds unless auth is enabled or
|
|
|
|
|
# TRUST_PROXY=true is set for a reverse-proxy deployment.
|
2026-03-19 19:00:24 -06:00
|
|
|
host: 127.0.0.1
|
|
|
|
|
|
2025-08-01 21:19:30 -04:00
|
|
|
# Port to listen on (default: 3001)
|
|
|
|
|
port: 3001
|
|
|
|
|
|
|
|
|
|
# Timeout configurations
|
|
|
|
|
timeouts:
|
|
|
|
|
# Maximum duration for reading the entire request, including the body
|
|
|
|
|
read: 10m
|
|
|
|
|
|
|
|
|
|
# Maximum duration before timing out writes of the response
|
|
|
|
|
write: 10m
|
|
|
|
|
|
|
|
|
|
# Maximum amount of time to wait for the next request when keep-alives are enabled
|
|
|
|
|
idle: 10m
|
|
|
|
|
|
|
|
|
|
# Provider configurations
|
|
|
|
|
providers:
|
|
|
|
|
# Anthropic Claude configuration
|
|
|
|
|
anthropic:
|
|
|
|
|
# Base URL for Anthropic API (can be changed for custom endpoints)
|
|
|
|
|
base_url: "https://api.anthropic.com"
|
2026-03-19 18:52:09 -06:00
|
|
|
|
2025-08-01 21:19:30 -04:00
|
|
|
# Maximum number of retries for failed requests
|
|
|
|
|
max_retries: 3
|
2026-03-19 18:52:09 -06:00
|
|
|
|
2025-08-01 21:19:30 -04:00
|
|
|
# OpenAI configuration
|
|
|
|
|
openai:
|
|
|
|
|
# API key for OpenAI
|
|
|
|
|
# Can also be set via OPENAI_API_KEY environment variable
|
2025-08-03 22:30:13 -04:00
|
|
|
# api_key: "..."
|
2026-03-19 18:52:09 -06:00
|
|
|
|
2025-08-01 21:19:30 -04:00
|
|
|
# Base URL for OpenAI API (can be changed for custom endpoints)
|
|
|
|
|
# Can also be set via OPENAI_BASE_URL environment variable
|
|
|
|
|
# base_url: "https://api.openai.com"
|
|
|
|
|
|
2026-03-19 18:52:09 -06:00
|
|
|
# Allow clients to provide their own API key via header
|
|
|
|
|
# Can also be set via OPENAI_ALLOW_CLIENT_API_KEY environment variable
|
|
|
|
|
allow_client_api_key: false
|
|
|
|
|
|
|
|
|
|
# Header name for client-provided API key (default: x-openai-api-key)
|
|
|
|
|
# Can also be set via OPENAI_CLIENT_API_KEY_HEADER environment variable
|
|
|
|
|
client_api_key_header: "x-openai-api-key"
|
|
|
|
|
|
|
|
|
|
# CORS Configuration
|
|
|
|
|
# Controls Cross-Origin Resource Sharing for the web UI
|
|
|
|
|
cors:
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# Allowed origins. The built-in defaults are permissive, so set these
|
|
|
|
|
# explicitly if you want tighter browser access.
|
2026-03-19 18:52:09 -06:00
|
|
|
# Can also be set via CORS_ALLOWED_ORIGINS environment variable (comma-separated)
|
|
|
|
|
allowed_origins:
|
2026-03-19 19:00:24 -06:00
|
|
|
- "http://localhost:3000"
|
|
|
|
|
- "http://127.0.0.1:3000"
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
- "http://localhost:5174"
|
|
|
|
|
- "http://127.0.0.1:5174"
|
2026-03-19 18:52:09 -06:00
|
|
|
|
|
|
|
|
# Allowed HTTP methods
|
|
|
|
|
# Can also be set via CORS_ALLOWED_METHODS environment variable (comma-separated)
|
|
|
|
|
allowed_methods:
|
|
|
|
|
- "GET"
|
|
|
|
|
- "POST"
|
|
|
|
|
- "DELETE"
|
|
|
|
|
- "OPTIONS"
|
|
|
|
|
|
2026-03-19 19:00:24 -06:00
|
|
|
# Allowed headers
|
2026-03-19 18:52:09 -06:00
|
|
|
# Can also be set via CORS_ALLOWED_HEADERS environment variable (comma-separated)
|
|
|
|
|
allowed_headers:
|
2026-03-19 19:00:24 -06:00
|
|
|
- "Accept"
|
|
|
|
|
- "Authorization"
|
|
|
|
|
- "Content-Type"
|
|
|
|
|
- "Anthropic-Version"
|
|
|
|
|
- "Anthropic-Beta"
|
|
|
|
|
- "X-API-Key"
|
|
|
|
|
- "X-Requested-With"
|
|
|
|
|
|
|
|
|
|
# Auth Configuration
|
|
|
|
|
# When enabled, all non-health endpoints require bearer token or X-API-Key auth.
|
|
|
|
|
auth:
|
|
|
|
|
# Enable auth for non-health endpoints
|
|
|
|
|
# Public/non-loopback binds must enable auth and set a token.
|
|
|
|
|
enabled: false
|
|
|
|
|
|
|
|
|
|
# Shared secret used for Authorization: Bearer <token> or X-API-Key: <token>
|
|
|
|
|
token: ""
|
|
|
|
|
|
|
|
|
|
# Header name used for API-key style auth
|
|
|
|
|
api_key_header: "x-api-key"
|
|
|
|
|
|
|
|
|
|
# Allow requests from localhost to bypass auth when enabled
|
|
|
|
|
allow_localhost_bypass: true
|
2026-03-19 18:52:09 -06:00
|
|
|
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# Optional dashboard-only password. When set, the Svelte dashboard and
|
|
|
|
|
# dashboard data endpoints require HTTP basic auth with username "admin".
|
|
|
|
|
dashboard_password: ""
|
|
|
|
|
|
|
|
|
|
# Set to true when running behind a trusted reverse proxy and you want to
|
|
|
|
|
# skip the public-bind auth requirement enforced at startup.
|
|
|
|
|
trust_proxy: false
|
|
|
|
|
|
2025-08-01 21:19:30 -04:00
|
|
|
# Storage configuration
|
|
|
|
|
storage:
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# Storage backend. Supported values: sqlite, postgres
|
|
|
|
|
db_type: "sqlite"
|
|
|
|
|
|
2025-08-01 21:19:30 -04:00
|
|
|
# SQLite database path for storing request history
|
|
|
|
|
db_path: "requests.db"
|
2026-03-19 19:00:24 -06:00
|
|
|
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# PostgreSQL connection string used when db_type=postgres
|
|
|
|
|
database_url: ""
|
|
|
|
|
|
2026-03-19 19:00:24 -06:00
|
|
|
# Keep request bodies in storage. Disable for metadata-only tracking.
|
|
|
|
|
capture_request_body: true
|
|
|
|
|
|
|
|
|
|
# Keep response bodies and streaming chunks in storage.
|
|
|
|
|
capture_response_body: true
|
|
|
|
|
|
|
|
|
|
# Store only request/response metadata, not payload bodies.
|
|
|
|
|
metadata_only: false
|
|
|
|
|
|
|
|
|
|
# Delete records older than this many days on write. 0 disables cleanup.
|
|
|
|
|
retention_days: 0
|
|
|
|
|
|
|
|
|
|
# JSON payload fields to redact before storage.
|
|
|
|
|
redacted_fields:
|
|
|
|
|
- api_key
|
|
|
|
|
- authorization
|
|
|
|
|
- token
|
|
|
|
|
- password
|
|
|
|
|
- secret
|
|
|
|
|
- access_token
|
|
|
|
|
- refresh_token
|
|
|
|
|
- client_secret
|
2025-08-01 21:19:30 -04:00
|
|
|
|
|
|
|
|
# Directory for storing request files (if needed in future)
|
|
|
|
|
# requests_dir: "./requests"
|
|
|
|
|
|
2025-08-03 22:30:13 -04:00
|
|
|
# Subagent Configuration (Optional)
|
|
|
|
|
# Enable this feature if you want to route specific Claude Code agents to different LLM providers
|
|
|
|
|
# For subagent setup instructions, see: https://docs.anthropic.com/en/docs/claude-code/sub-agents
|
2025-08-01 21:19:30 -04:00
|
|
|
subagents:
|
2025-08-03 22:30:13 -04:00
|
|
|
# Enable subagent routing (default: false)
|
|
|
|
|
enable: false
|
|
|
|
|
|
|
|
|
|
# Maps subagent types to specific models
|
|
|
|
|
# Only used when enable: true
|
2025-08-01 21:19:30 -04:00
|
|
|
mappings:
|
|
|
|
|
# Code review specialist (example)
|
|
|
|
|
# code-reviewer: "gpt-4o"
|
|
|
|
|
|
|
|
|
|
# Data analysis expert (example)
|
2025-08-03 22:30:13 -04:00
|
|
|
# data-analyst: "o3"
|
2025-08-01 21:19:30 -04:00
|
|
|
|
|
|
|
|
# Documentation writer (example)
|
|
|
|
|
# doc-writer: "gpt-3.5-turbo"
|
|
|
|
|
|
|
|
|
|
# Environment variable overrides:
|
|
|
|
|
# The following environment variables will override the YAML configuration:
|
|
|
|
|
#
|
|
|
|
|
# Server:
|
2026-03-19 19:00:24 -06:00
|
|
|
# SERVER_HOST - Bind host (default: 127.0.0.1)
|
2026-03-19 18:52:09 -06:00
|
|
|
# PORT - Server port
|
|
|
|
|
# READ_TIMEOUT - Read timeout duration
|
|
|
|
|
# WRITE_TIMEOUT - Write timeout duration
|
|
|
|
|
# IDLE_TIMEOUT - Idle timeout duration
|
2025-08-01 21:19:30 -04:00
|
|
|
#
|
|
|
|
|
# Anthropic:
|
2026-03-19 18:52:09 -06:00
|
|
|
# ANTHROPIC_FORWARD_URL - Anthropic base URL
|
|
|
|
|
# ANTHROPIC_VERSION - Anthropic API version
|
|
|
|
|
# ANTHROPIC_MAX_RETRIES - Maximum retries for Anthropic requests
|
2025-08-01 21:19:30 -04:00
|
|
|
#
|
|
|
|
|
# OpenAI:
|
2026-03-19 18:52:09 -06:00
|
|
|
# OPENAI_API_KEY - OpenAI API key
|
|
|
|
|
# OPENAI_BASE_URL - OpenAI base URL
|
|
|
|
|
# OPENAI_ALLOW_CLIENT_API_KEY - Allow client-provided API keys (true/false)
|
|
|
|
|
# OPENAI_CLIENT_API_KEY_HEADER - Header name for client API key
|
2025-08-01 21:19:30 -04:00
|
|
|
#
|
2026-03-19 19:00:24 -06:00
|
|
|
# Auth:
|
|
|
|
|
# AUTH_ENABLED - Enable auth for non-health endpoints (true/false)
|
|
|
|
|
# AUTH_TOKEN - Shared secret for bearer / API-key auth
|
|
|
|
|
# AUTH_API_KEY_HEADER - Header name for API-key style auth
|
|
|
|
|
# AUTH_ALLOW_LOCALHOST_BYPASS - Allow loopback requests to bypass auth (true/false)
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# DASHBOARD_PASSWORD - Dashboard HTTP basic auth password
|
|
|
|
|
# TRUST_PROXY - Skip public-bind auth enforcement behind a reverse proxy
|
2026-03-19 19:00:24 -06:00
|
|
|
#
|
2025-08-01 21:19:30 -04:00
|
|
|
# Storage:
|
Local fork: hardening + ops improvements (timeout knob, demotion, /livez, drain)
This commit captures both the prior accumulated work-in-progress
(framework migration web/→svelte/, postgres storage, conversation
viewer, dashboard auth, OpenAPI spec, integration tests) AND today's
operational improvements layered on top. History wasn't checkpointed
incrementally; happy to split it via interactive rebase if a reviewer
wants smaller commits.
Today's changes (in addition to the older WIP):
1. Configurable upstream response-header timeout
- ANTHROPIC_RESPONSE_HEADER_TIMEOUT env (default 300s)
- Replaces hardcoded 300s in provider/anthropic.go that was firing
on opus + 1M-context + extended thinking non-streaming requests
- Files: internal/config/config.go, internal/provider/anthropic.go
2. Structured forward-error diagnostic logging
- When a forward to Anthropic fails, log a single key=value line
with request_id, model, stream, body_bytes, has_thinking,
anthropic_beta, query, elapsed, ctx_err — alongside the existing
human-readable error line for back-compat
- Files: internal/handler/handlers.go (logForwardFailure)
3. Full SSE protocol passthrough + Flusher fix
- handler/handlers.go: forward all SSE lines verbatim (event:, id:,
retry:, : comments, blank-line terminators), not only data:.
Previous code produced malformed SSE for strict parsers.
- middleware/logging.go: explicit Flush() method on responseWriter.
Embedding http.ResponseWriter (interface) does not auto-promote
Flush(), so every w.(http.Flusher) check in the streaming
handler was returning ok=false and SSE writes buffered in net/http
until the body closed.
4. Non-streaming → streaming demotion (feature-flagged)
- ANTHROPIC_DEMOTE_NONSTREAMING env (default false)
- When enabled and the routed provider is anthropic, force stream=true
upstream for clients that asked for stream=false. Receive SSE,
accumulate via accumulateSSEToMessage (handles text, tool_use with
partial_json reassembly, thinking, signature, citations_delta,
usage merge), and synthesize a single non-streaming JSON response.
- Eliminates the ResponseHeaderTimeout class of failure entirely.
- Body rewrite uses json.Decoder + UseNumber() to preserve integer
precision in unknown nested fields (tool inputs from prior turns).
- Files: internal/config/config.go, internal/handler/handlers.go,
cmd/proxy/main.go, cmd/proxy/main_test.go
5. Live operational state: /livez gauge + graceful drain
- New internal/runtime package: atomic in-flight counter + draining flag
- New middleware/inflight.go: increments runtime gauge, applied to
/v1/* subrouter so Messages, ChatCompletions, and ProxyPassthrough
are all counted
- /v1/* moved to a gorilla/mux subrouter so the InFlight middleware
applies surgically; /health, /livez, /openapi.* remain on parent
router (unauthenticated, uncounted)
- Health handler returns 503 draining when runtime.IsDraining() is
true, so Traefik stops routing to a slot before drain begins
- New /livez handler returns {status, in_flight, draining, timestamp}
- SIGTERM handler in main.go: SetDraining(true), poll for in_flight==0
with 32-min ceiling and 1s tick (logs every 10s), then srv.Shutdown
- Auth bypass list extended with /livez
- Files: internal/runtime/runtime.go (new),
internal/middleware/inflight.go (new),
internal/middleware/auth.go,
internal/handler/handlers.go (Health, Livez, runtime import),
cmd/proxy/main.go (subrouter, drain loop)
6. OpenAPI spec updates
- Document Health 503 response and new DrainingResponse schema
- Add /livez path with LivezResponse schema
- Files: internal/handler/openapi.go
Verified: go build ./... clean, go test ./... all pass, go vet clean.
Three rounds of codex peer review across changes 1-5; all feedback
addressed (citations_delta, json.Number precision, drain-loop logging
via lastLog timestamp, PathPrefix tightened to "/v1/").
2026-05-02 15:15:58 -06:00
|
|
|
# DB_TYPE - Storage backend (sqlite/postgres)
|
|
|
|
|
# DATABASE_URL - PostgreSQL connection string
|
2026-03-19 18:52:09 -06:00
|
|
|
# DB_PATH - Database file path
|
2026-03-19 19:00:24 -06:00
|
|
|
# STORAGE_CAPTURE_REQUEST_BODY - Keep request bodies (true/false)
|
|
|
|
|
# STORAGE_CAPTURE_RESPONSE_BODY - Keep response bodies (true/false)
|
|
|
|
|
# STORAGE_METADATA_ONLY - Store metadata only (true/false)
|
|
|
|
|
# STORAGE_RETENTION_DAYS - Delete rows older than N days
|
|
|
|
|
# STORAGE_REDACTED_FIELDS - Comma-separated payload fields to redact
|
2026-03-19 18:52:09 -06:00
|
|
|
#
|
|
|
|
|
# CORS:
|
|
|
|
|
# CORS_ALLOWED_ORIGINS - Comma-separated allowed origins
|
|
|
|
|
# CORS_ALLOWED_METHODS - Comma-separated allowed methods
|
|
|
|
|
# CORS_ALLOWED_HEADERS - Comma-separated allowed headers
|
2025-08-01 21:19:30 -04:00
|
|
|
#
|
|
|
|
|
# Subagents:
|
2026-03-19 18:52:09 -06:00
|
|
|
# SUBAGENT_MAPPINGS - Comma-separated subagent:model pairs
|
2026-03-19 19:00:24 -06:00
|
|
|
# Example: "code-reviewer:claude-3-5-sonnet"
|