copilot-relay 0.3.5 → 0.3.7

package/README.md

@@ -75,7 +75,17 @@ With `claudeSetup: true`, `start` manages `ANTHROPIC_BASE_URL`, a dummy
 `ANTHROPIC_AUTH_TOKEN` when auth is absent, and the configured GPT default via
 the top-level `model` field in `~/.claude/settings.json`. Exact
 `gpt-6-astra` and `gpt-5.6-sol` overrides are normalized to Claude-facing
-`[1m]` identities; unrelated model choices are preserved.
+`[1m]` identities when discovery reports exactly 1M; other reported windows use
+plain GPT IDs with Claude's numeric context setting. Unrelated model choices and
+existing token-budget overrides are preserved.
+
+Startup retains Copilot's actual context, prompt, and output limits and exposes
+them on `GET /v1/models`. Managed Claude setup seeds
+`CLAUDE_CODE_MAX_CONTEXT_TOKENS` and `CLAUDE_CODE_MAX_OUTPUT_TOKENS` when absent.
+The default models support 1M total context, with up to 128K output tokens for
+Astra and 64K for Opus. Input text is never shortened; smaller explicit output
+budgets remain respected. See [full-context configuration](wiki/EN-Configuration.md)
+for existing/manual Claude setups and the limits on input plus output.
 
 ## Config
 
@@ -113,10 +123,16 @@ for live model discovery, compatible thinking efforts, and copy-paste updates.
 Any other `logLevel` value is invalid and stops startup.
 
 Valid `thinkEffort`: `none`, `low`, `medium`, `high`, `xhigh`, `max`.
+It is a fallback, not a forced override: Claude Code's `output_config.effort`
+wins, followed by the legacy `reasoning_effort` field, then `thinkEffort` when
+neither request field supplies a value. The selected effort is preserved through
+all model and WebSearch passes; existing config values are not rewritten.
 
 `upstreamTimeoutSeconds` controls the maximum time a single Claude request can
 spend waiting on upstream Copilot calls, including chat, Responses, preflight,
-and bridge-managed WebSearch calls. The default is `180`.
+and bridge-managed WebSearch calls. The default is `180`. Set `0` to disable the
+relay deadline for very long requests; client cancellation and provider/transport
+timeouts still apply. Existing saved timeouts are never changed by an upgrade.
 
 `webSearchBackend` controls bridge-managed Claude WebSearch. Leave it empty to
 use `gptModel`, or set a Copilot Responses model ID such as `gpt-5.5`.

package/config.default.yaml

@@ -21,6 +21,7 @@ copilotBaseUrl: https://api.githubcopilot.com
 
 # When true, start updates ~/.claude/settings.json with ANTHROPIC_BASE_URL
 # and a dummy ANTHROPIC_AUTH_TOKEN so Claude Code can use this proxy.
+# Also seeds context/output budgets from model discovery when not already set.
 claudeSetup: true
 
 # Log verbosity:
@@ -38,11 +39,14 @@ logLevel: info
 # midnight, which is what makes this setting take effect at all.
 logRetentionDays: 3
 
-# Default upstream thinking/reasoning effort sent to Copilot.
+# Fallback upstream effort when Claude Code does not specify one.
+# Request output_config.effort wins, followed by legacy reasoning_effort.
 # Valid values: none, low, medium, high, xhigh, max.
 thinkEffort: max
 
 # Max seconds to wait for one Claude request's upstream Copilot calls.
+# Set 0 to disable the relay deadline for very long requests. Client cancellation
+# and upstream/provider limits still apply.
 upstreamTimeoutSeconds: 180
 
 # Copilot model used for bridge-managed Claude WebSearch.