@hsupu/copilot-api 0.7.19 → 0.7.21

package/config.example.yaml

@@ -2,6 +2,21 @@
 # Copy this file to ~/.local/share/copilot-api/config.yaml and customize as needed.
 # All settings are hot-reloadable unless noted otherwise.
 
+# ============================================================================
+# Proxy
+# ============================================================================
+# Proxy URL for all outgoing requests to GitHub / Copilot APIs.
+# Supports http://, https://, socks5://, socks5h:// schemes.
+# socks5h:// routes DNS through the proxy (recommended for privacy).
+# Authentication via URL credentials: socks5h://user:pass@host:port
+# Takes precedence over HTTP_PROXY/HTTPS_PROXY environment variables.
+# CLI --proxy flag takes precedence over this setting.
+# NOT hot-reloadable (requires restart).
+
+# proxy: "http://127.0.0.1:7890"
+# proxy: "socks5h://127.0.0.1:1080"
+# proxy: "socks5h://user:pass@proxy.example.com:1080"
+
 # ============================================================================
 # Model
 # ============================================================================
@@ -33,21 +48,6 @@ model_overrides:
   claude-sonnet-4.5: claude-sonnet-4.6  # Latest sonnet
   claude-haiku-4.5: claude-sonnet-4.5   # Upgrade haiku to sonnet
 
-# ============================================================================
-# Proxy
-# ============================================================================
-# Proxy URL for all outgoing requests to GitHub / Copilot APIs.
-# Supports http://, https://, socks5://, socks5h:// schemes.
-# socks5h:// routes DNS through the proxy (recommended for privacy).
-# Authentication via URL credentials: socks5h://user:pass@host:port
-# Takes precedence over HTTP_PROXY/HTTPS_PROXY environment variables.
-# CLI --proxy flag takes precedence over this setting.
-# NOT hot-reloadable (requires restart).
-
-# proxy: "http://127.0.0.1:7890"
-# proxy: "socks5h://127.0.0.1:1080"
-# proxy: "socks5h://user:pass@proxy.example.com:1080"
-
 # ============================================================================
 # Timeouts
 # ============================================================================
@@ -56,7 +56,7 @@ model_overrides:
 stream_idle_timeout: 300       # Max seconds between SSE events (0 = no timeout).
                                # Applies to all streaming paths (Anthropic, Chat Completions, Responses).
 
-fetch_timeout: 300             # Seconds: request start → HTTP response headers (0 = no timeout).
+fetch_timeout: 600             # Seconds: request start → HTTP response headers (0 = no timeout).
                                # Applies to all upstream API clients.
 
 stale_request_max_age: 600     # Max seconds an active request can live before the stale reaper
@@ -64,17 +64,22 @@ stale_request_max_age: 600     # Max seconds an active request can live before t
                                # Safety net for requests that never complete/fail normally.
 
 # ============================================================================
-# Rate Limiter
+# Shutdown
 # ============================================================================
-# Fine-tune the adaptive rate limiter behavior.
-# The rate limiter is enabled/disabled via --no-rate-limit CLI flag;
-# these settings control its parameters when enabled.
+# Control graceful shutdown timing.
 
-rate_limiter:
-  retry_interval: 10        # Seconds to wait before retrying after rate limit error
-  request_interval: 10      # Seconds between requests in rate-limited mode
-  recovery_timeout: 10      # Minutes before attempting recovery from rate-limited mode
-  consecutive_successes: 5   # Consecutive successes needed to exit rate-limited mode
+shutdown:
+  graceful_wait: 300  # Phase 2: seconds to wait for in-flight requests to complete naturally (default: 60)
+  abort_wait: 600     # Phase 3: seconds to wait after abort signal for handlers to wrap up (default: 120)
+
+# ============================================================================
+# History
+# ============================================================================
+# Control history recording behavior.
+
+history:
+  limit: 200          # Maximum number of entries to keep in memory (0 = unlimited, default: 200)
+  min_entries: 50     # Minimum entries to keep even under memory pressure (default: 50)
 
 # ============================================================================
 # Anthropic
@@ -82,18 +87,39 @@ rate_limiter:
 # Settings for Anthropic API tool handling and timeouts.
 
 anthropic:
-  convert_server_tools_to_custom: true  # Convert server-side tools (web_search) to custom tool format
-  dedup_tool_calls: false    # false | "input" | "result" (true = "input" for compat)
-                             # "input": dedup by (name, input); "result": also require identical result
-  truncate_read_tool_result: false  # Strip <system-reminder> tags from Read tool results
-  # rewrite_system_reminders: false   # false = keep all (default), true = remove all
-  rewrite_system_reminders:           # Or provide rewrite rules (first match wins, top-down).
-                                      # Note: `model` field is NOT supported here (only in system_prompt_overrides).
+  strip_server_tools: false               # Strip server-side tools (web_search, etc.) from requests
+  dedup_tool_calls: false                 # false | "input" | "result" (true = "input" for compat)
+                                          # "input": dedup by (name, input); "result": also require identical result
+  strip_read_tool_result_tags: false      # Strip <system-reminder> tags from Read tool results
+  context_editing: off                    # off | clear-thinking | clear-tooluse | clear-both
+                                          # Server-side context editing mode. Controls how Anthropic's
+                                          # context_management trims older context when input grows large.
+                                          # - off: disabled (default). No context_management sent.
+                                          # - clear-thinking: clear old thinking blocks.
+                                          # - clear-tooluse: clear old tool_use/result pairs.
+                                          # - clear-both: apply both clear-thinking and clear-tooluse.
+                                          # Only effective for models supporting context editing.
+  # rewrite_system_reminders: false       # false = keep all (default), true = remove all
+  rewrite_system_reminders:               # Or provide rewrite rules (first match wins, top-down).
+                                          # Note: `model` field is NOT supported here (only in system_prompt_overrides).
     - from: "^Whenever you read a file, you should consider whether it would be considered malware"
       to: ""                              # Empty = remove the tag
     # - from: ".*"                        # Catch-all: keep unchanged (gms flags are automatic)
     #   to: "$0"                          # $0 = original content
 
+# ============================================================================
+# Rate Limiter
+# ============================================================================
+# Fine-tune the adaptive rate limiter behavior.
+# The rate limiter is enabled/disabled via --no-rate-limit CLI flag;
+# these settings control its parameters when enabled.
+
+rate_limiter:
+  retry_interval: 10        # Seconds to wait before retrying after rate limit error
+  request_interval: 10      # Seconds between requests in rate-limited mode
+  recovery_timeout: 10      # Minutes before attempting recovery from rate-limited mode
+  consecutive_successes: 5   # Consecutive successes needed to exit rate-limited mode
+
 # ============================================================================
 # Auto-Truncate
 # ============================================================================
@@ -104,24 +130,6 @@ anthropic:
 # context size before resorting to message removal. Default: true.
 # compress_tool_results_before_truncate: false
 
-# ============================================================================
-# History
-# ============================================================================
-# Control history recording behavior.
-
-# Maximum number of history entries to keep in memory.
-# 0 = unlimited. Default: 200.
-history_limit: 200
-
-# ============================================================================
-# Shutdown
-# ============================================================================
-# Control graceful shutdown timing.
-
-shutdown:
-  graceful_wait: 60   # Phase 2: seconds to wait for in-flight requests to complete naturally (default: 60)
-  abort_wait: 120     # Phase 3: seconds to wait after abort signal for handlers to wrap up (default: 120)
-
 # ============================================================================
 # System Prompt
 # ============================================================================