v0.3.1

fix(core): normalize model IDs for context limit and pricing lookup

Dot-keyed entries (e.g. claude-opus-4.6) were not matching hyphenated API model IDs (e.g. claude-opus-4-6-20251101), causing fallthrough to the shorter claude-opus-4 key. Opus 4.6 was showing 200k context instead of 1M.

Normalizes dots to hyphens in both the model string and lookup key before matching in getContextLimit() and estimateCost().

v0.3.0

@contextio/core@0.3.0

• Add CREDENTIAL_PATTERNS export — canonical detection regexes for private keys, AWS, GitHub, Anthropic, OpenAI, and generic secret assignments. Single source of truth shared with @contextio/redact.
• Fix scanRegex required mode: isSafe now correctly returns true only when zero alerts are raised, regardless of mode.
• Add Gemini Code Assist nested .response wrapper support in parseResponseUsage.
• Improve JSDoc across output-scanner, output-urls, response, and security modules.

@contextio/redact@0.3.0

• Derive AWS key, GitHub, Anthropic, OpenAI, and generic-secret redaction rules from @contextio/core CREDENTIAL_PATTERNS via toRule(). Eliminates duplicated regex literals between detection and redaction.

@contextio/proxy@0.2.3

• Documentation improvements to forward.ts internals.

@contextio/cli@0.3.0

• Remove replay and export commands (incomplete stubs).
• Read CLI version from package.json instead of hardcoded string.
• Replace Record<string, any> with JsonObject + asObj() helper in inspect.ts for safer field access.

@contextio/core v0.2.3

Bug Fixes

• core: Correct Anthropic cache write pricing multiplier from 0.25x to 1.25x (#29)
• routing: Normalize bare /responses to /v1/responses for Codex Enterprise

v0.2.3

Update models: newer Claude/GPT-5.x/Gemini entries, fix o1-mini pricing ([3,12] → [1.1,4.4]), add Gemini cache pricing

v0.2.2

Fix /codex/ path routing to chatgpt upstream and double /v1 in OpenAI upstream (proxy package).

v0.2.1

Fix /codex/ path routing to chatgpt upstream for Pi openai-codex provider. Fix double /v1 in OpenAI upstream default.

v0.2.0

What's new

Vertex AI support

• New vertex provider in routing: classifies Vertex AI paths and resolves per-location upstream URLs
• UPSTREAM_VERTEX_URL env var for custom override (defaults to us-central1-aiplatform.googleapis.com)
• ResolveTargetResult now includes apiFormat so callers no longer need a second classify call

International PII redaction

• pii preset: EU phone numbers (+31/+32/+33/+34/+39/+41/+43/+44/+45/+46/+47/+48/+49) and IBAN
• strict preset: Dutch BSN, UK National Insurance number, passport numbers
• All new rules are context-guarded to minimise false positives
• Full reversible redaction round-trip support for all new types

Docker

• Official Docker image published to GHCR
• Multi-arch build (amd64 + arm64)
• See docker/README.md for usage

v0.1.1

Fix npm install for published packages.

v0.1.0 was published with unresolved workspace:* dependencies. This release fixes that so npm install -g @contextio/cli works correctly.

v0.1.0

First public release.

• Local proxy for AI coding tools with logging and PII redaction
• Reversible redaction mode (strip on request, restore on response)
• Support for Claude CLI, Gemini CLI, Aider, Pi, Codex, OpenCode, Copilot
• CLI commands: proxy, attach, monitor, inspect, replay, export, doctor
• Three built-in redaction presets (secrets, pii, strict) plus custom policies
• Zero runtime dependencies in the proxy trust path