v0.2.0

Changelog

• ff4feeb: security: harden transport, storage, redaction, and CI (@gl0bal01)

v0.1.2

Changelog

• 26e5738: fix: harden doctor live ping and SSE transport — release v0.1.2 (@gl0bal01)
• ec32bcd: style: drop trailing newline from internal/api/client.go (@gl0bal01)

v0.1.1

Fixed

• omi consensus: per-model answers no longer corrupted by mid-stream error text. consensusAnswer now writes stderr to io.Discard instead of the same buffer that captures the answer fed to the synthesis model.
• Timeout error message now reflects the configured --timeout value (or api.DefaultTimeout) instead of a hardcoded 60s literal.
• DeleteConversation wraps the conversation UUID with url.PathEscape so an unexpected character cannot reshape the request path.

Changed

• Transcribe cascade reduced from three POSTs (audioUrl → audio → path) to two (audioUrl → audio). One fallback is sufficient for backend-field compatibility, and /api/features is non-idempotent so additional retries only enlarge the failure surface.
• Speech-to-text model metadata is now a single transcribeEntries registry plus rawTranscribeModels for ids without an alias. The legacy slice and map are derived at package init from this source of truth.
• Secret masking is centralized in a new internal/redact package. internal/api now calls redact.APIKey instead of its own maskSecret; config.MaskAPIKey delegates to the same implementation.
• REPL bufio.Scanner initial buffer raised from 1 KiB to 64 KiB so multi-line paste avoids several immediate grow cycles. The 1 MiB ceiling is unchanged.

v0.1.0

Added

• Initial release of omi, a Go CLI for the 1min.ai REST API.
• Streaming chat via POST /api/chat-with-ai with UNIFY_CHAT_WITH_AI.
• Vision and document Q&A via the same unified endpoint with imageList and files.
• Code generation via POST /api/features with CODE_GENERATOR.
• Audio transcription via POST /api/features with SPEECH_TO_TEXT.
• Asset upload via POST /api/assets (multipart).
• Server-side conversation lifecycle: create on first turn, DELETE on omi session clear (idempotent: 404 treated as success, 401 surfaces as hard error).
• Named sessions persisted at ~/.config/omi/sessions.json.
• Embedded model registry with capability-typed aliases (chat / code-only / vision-only) and per-model defaults (webSearch, numOfSite, maxWord, conversationType).
• ~/.config/omi/models.json override (full-replacement semantics).
• Shell completion for bash, zsh, fish, and powershell.
• REPL when stdin is a TTY; one-shot when stdin is piped.
• Friendly error UX with consistent omi: <msg> format and exit codes 0/1/2.
• omi consensus, a client-side 3-model panel feature that asks multiple chat models and synthesizes consensus, disagreements, and recommendation. Panel fans out concurrently; wall-clock latency is max(per-model) not sum(per-model). Answer order matches the -m input order; first error cancels siblings.
• --api-key flag (priority: flag → OMI_API_KEY env → config file).
• --timeout flag for non-streaming HTTP calls (default 60s).
• HTTP retry layer honoring Retry-After (RFC 7231 seconds or HTTP-date), capped at 30s, and respecting request-context cancellation.
• Help, quickstart, and README examples for consensus, sessions, and mixed-history usage.
• Repeatable quality gates for fmt, unit tests, race tests, lint, security scanning, vuln scanning, smoke, and full verification.
• CI enforcement for formatting, unit tests, race tests, lint, gosec, govulncheck, builds, and binary smoke.
• Release checklist with versioning, smoke, artifact, and checksum verification steps.
• Automation documentation for stable --json outputs.

Security

• Sanitized streamed content and streamed API error messages to prevent terminal control-sequence injection.
• Session persistence permission tests for 0700 config directories and 0600 session files.
• SIGINT/SIGTERM cancel in-flight HTTP and SSE work; clean cancel exits with status 130 (no "context canceled" noise on stderr).
• Capped --no-stream buffered output at 16 MiB with a single stderr truncation warning to bound memory under runaway upstream responses.
• Capped piped-stdin prompt reads at 4 MiB; oversize input fails fast with a usage error.
• Capped non-streaming success bodies (code, transcribe) at 16 MiB and upstream error bodies at 64 KiB to bound log/error surface from a hostile or runaway server.

Performance

• Cold-start measurement (linux/amd64, Go 1.25): ~6.07ms median of 10 runs
  of ./bin/omi --version (-ldflags "-s -w"). Well under the <20ms target.

  for i in $(seq 1 10); do /usr/bin/time -f '%e' ./bin/omi --version 2>&1 >/dev/null; done | sort -n | sed -n '5p'