v0.3.1

0.3.1 — 2026-10-02

System One replay between TypeSafe and Ollama, a tidier Compare view, IBM Plex Sans, and winget.

Added

• System One decisions replay between TypeSafe and Ollama 0.35, which serves the same
  POST /v1/systemone API: a typesafe-systemone capture can replay to any ollama backend
  or to an api.typesafe.ai backend. Credentials come only from the target, so a TypeSafe key
  never reaches Ollama. OpenRouter's /api/alpha/decisions stays same-backend only (#137).
• Package-manager install on Windows with winget: winget install spenceclark.Vessel (#32).

Changed

• Compare hides metric rows that no column has a value for, in every format — for example
  TTFT, Tok/s and Stop reason for System One (#137).
• For a decision pair, Compare's shared request shows only the parameter diff and state;
  each answer column already carries its questions (#137).
• Decision values, confidence and the score scale are shown rounded to 3 places; Raw JSON
  keeps full precision (#137).
• Choice bars follow the request's criteria order, so Compare columns line up. In the request
  detail view the winning option is no longer always the top bar (#137).
• The UI font is now IBM Plex Sans, bundled with the app. The previous Inter font never
  loaded because of a font-name mismatch, so the UI fell back to the system font (#127).
• Dependency bumps: @tanstack/react-query 5.104.0, @tanstack/react-virtual 3.14.13 (#128, #131).

v0.3.0

0.3.0 — 2026-09-19

Tools tab, Anthropic server tools, TypeSafe System One, MCP resources, and two new warnings.

Added

• Tools tab: a request that declares tools gets a Tools (n) tab with one card per tool —
  description, parameter table, and how many times it was called (#86).
• Anthropic server-tool blocks (server_tool_use, web search and web fetch results) render
  as cards, with numbered citations and a sources list per message (#88).
• Tool call args and tool results that parse to an object render as a readable field list,
  with multi-line strings unescaped; Raw JSON still shows the exact bytes (#91).
• TypeSafe System One (Jev) support: typesafe-systemone traffic — direct /v1/systemone or
  OpenRouter's /api/alpha/decisions — is detected instead of falling to raw, with a
  Decisions view (one card per question, probability bars) and same-backend replay (#122).
• Two new warnings: repetitive_output for a response stuck in a loop, and
  slow_response for a non-streamed request slower than warnings.slowResponseMs
  (default 120s, 0 disables; editable in the Config panel) (#90).
• MCP resources: vessel://requests/{id} and vessel://sessions/{id}, with
  resources/list returning the most recent requests and sessions (#92).
• OpenRouter and TypeSafe in the "Add backend" catalog (#124).

Fixed

• A queued replay whose backend was repointed while it waited is refused with
  409 replay_target_changed instead of sending the old credential to the new
  destination (#106).
• MCP request reads decode bodies under capture.maxBodyMb rather than unbounded, and
  report truncation with decodeTruncated (#107).
• While a listen-address change is pending restart, the UI/API stays reachable on the
  address the process is actually bound to (#112).
• path_missing_v1 only fires for the missing-/v1 mistake: an OpenAI SDK path suffix, no
  /v1/ segment anywhere, and a backend baseUrl with no path of its own (#123).
• Anthropic server-tool blocks are included in MCP include=text and full-text search;
  encrypted_* fields never reach the flattened text (#89).
• openai-chat reasoning text under the reasoning key (e.g. Ollama) is no longer dropped
  from the thinking block and search (#80).
• Anthropic rate-limit headers are parsed (#108).
• Parameter sweeps on OpenAI Responses replays use max_output_tokens (#111).
• Config validation rejects a backend with "type": null (#109).
• Multi-replay rejects a null variation with invalid_request and the offending index,
  before dispatching anything (#110).

Changed

• Dependency bumps: react 19.3.0, tailwind-merge 3.7.0 (#117, #119).

v0.2.1

0.2.1 — 2026-09-13

Post-0.2.0 fixes.

Fixed

• Ctrl+C no longer hangs on shutdown, and the startup console output is clearer (#64).
• The live request feed reconnects when the browser's EventSource enters CLOSED instead
  of silently going stale (#75).
• A backend connection failure logs one clean warning line instead of a YARP stack
  trace (#76).

Changed

• Dependency bumps: Microsoft.Data.Sqlite 10.0.12, lucide-react 1.43.0,
  @tanstack/react-virtual 3.14.11 (#68, #69, #72).

v0.2.0

0.2.0 — 2026-09-07

Named sessions, multi-replay with scoring, reports, export, and package-manager installs.

Added

• Named sessions: assign requests to a run with X-Vessel-Session (created on first use),
  with a session picker, per-session stats, and session deletion (#29, #41).
• Multi-replay: fan one captured request across several models, or one model across
  several parameter values, and see every response side by side with per-column metric
  deltas (#48).
• Scoring: rate any response 1–5 from the Compare view — original included — and a
  leaderboard in Reports ranks models and parameter sets by mean score and win-rate (#49).
• Reports view: requests, tokens, tok/s, duration, cache efficiency, and warnings by
  model/tag/backend, plus a context-growth chart per session or tag (#25, #26).
• Export the current filtered request list to CSV or JSONL (#24).
• Replay dialect fix-ups: max_tokens ↔ max_completion_tokens is renamed automatically
  for OpenAI Chat replays and shown as (auto) in Compare's parameter diff (#28).
• Package-manager installs: Homebrew (macOS and Linux) and Scoop (#31, #32).
• Tool calls a model emits as plain text instead of structured output are detected and
  flagged (#40).
• A hint on 404s from OpenAI-compatible backends when the client's base_url is missing
  /v1 (#57).
• README: tagging agents and naming runs from LangChain / LangGraph.

Changed

• Compare is now a grid rather than a fixed pair (a pair is the two-column case); metrics
  render as a table across columns (#48).
• MCP search_requests and get_request include each request's score and replay linkage
  (#48, #49).
• Config panel: backend cards compacted to two rows with shared explainers (#42).
• macOS unblock steps updated for Sequoia (#39).

The database schema migrates automatically on first start; the additions are additive, so
an older Vessel can still open the file.

v0.1.1

Changelog

All notable changes are documented here. Vessel follows Semantic Versioning.

0.1.1 — 2026-08-30

Post-launch fixes and polish.

Fixed

• Copy-as-curl now emits the correct auth for Anthropic backends (x-api-key +
  anthropic-version) rather than an OpenAI Bearer header — including backends left on
  type: auto (#7).
• Copy-as-curl targets a reachable host instead of the bind address, so a command copied
  while Vessel runs in a container actually connects (#8).
• Chrome DevTools' /.well-known/appspecific/… probe is no longer proxied and captured as
  an errored request (#4).

Added

• Config validation rejects http:// for public/remote backends, so an API key is never
  sent in plaintext; localhost and LAN backends are unaffected (#5).
• Keyboard navigation — ↑/↓ move through the request list (#6).
• "Add backend" is now a picker of known backends (OpenAI, Anthropic, Ollama, vLLM,
  llama.cpp, LM Studio, …) that prefills URL, type, and auth env var (#9).
• First-run guidance when Ollama isn't reachable, pointing you to add a backend (#11).

Changed

• Clearer label and help text for the streamed-token-usage setting (#10).

0.1.0 — 2026-08-29

First public release of Vessel: a local-first, single-binary reverse proxy for LLM
traffic. It captures and searches OpenAI, Anthropic, Ollama, and raw traffic; provides
live metrics, replay and Compare; includes a read-only MCP endpoint; and ships native
Windows, Linux, and macOS archives plus a GHCR container image.

v0.1.0

Changelog

All notable changes are documented here. Vessel follows Semantic Versioning.

0.1.0 — 2026-08-29

First public release of Vessel: a local-first, single-binary reverse proxy for LLM
traffic. It captures and searches OpenAI, Anthropic, Ollama, and raw traffic; provides
live metrics, replay and Compare; includes a read-only MCP endpoint; and ships native
Windows, Linux, and macOS archives plus a GHCR container image.