v0.5.0

Changelog v0.5.0

This version doesn't have much new functionality, but it features the new ConversationStore (experimental) and a lot of restructuring and cleanup, paving the way for new features in future versions such as #47.

Breaking Changes

Vault format changed

Please re-authenticate after upgrading.

Renamed config options

• conversations.sync.projectName → conversations.projectName
• conversations.sync.enabled → conversations.enableSync

Removed config options

• conversations.sync.projectId: use conversations.projectName instead
• conversations.sync.includeSystemMessages
• conversations.sync.autoSync and /sync chat command
• conversations.maxInMemory

Major Features

Upstream Conversation Store Integration (PR #58)

• Added persistent conversation storage using Proton's Redux + IndexedDB + Sagas infrastructure.
• Added useFallbackStore config option (default: true) for gradual migration

Package Restructuring

• Reorganized upstream Proton code into npm workspace packages:
  • packages/lumo/: synced from https://github.com/ProtonMail/WebClients/applications/lumo/src/app/
  • packages/proton/: synced from https://github.com/ProtonMail/WebClients/packages/
• Established patch strategy: "pull unchanged > patch > shim"
• Simplified sync-upstream workflow with better tracking

Native Tool Call Persistence (PR #59)

• Web search, weather, stock, and crypto tool calls now persist correctly

Bug Fixes

• Fixed invalid conversation continuations when API client changes tool call outputs (#52)
• Fixed inconsistencies between streamed and persisted messages
• Fixed tool call deduplication (don't drop semanticId)
• Fixed config.yaml saving issues (#53)
• Fixed rclone auth not showing messages on text input
• Fixed npm build not using project-local tsc
• Fixed Docker build dependencies and missing packages/ copy (#61)
• Fixed Lumo tool call format: normalize nested arguments.parameters
• Fixed README to avoid launching docker without config.yaml

Improvements

Auth Refactoring

• Unified vault structure and auth providers
• Login/rclone auth now generates fallback keys if no keys available
• Browser auth decrypts session blob upon extraction and only saves keys

Config & Error Handling

• Improved config file handling: proper directory checks, valid scheme validation, missing auth section handling
• Better error messages for vault/keyfile misconfiguration
• Hide stack traces for known errors and invalid config values
• Show informative errors when db file can't be accessed

CLI & Commands

• /save now saves current conversation only (#41)
• Changed obsolete "npm run auth" to "tamer auth" in docs & messages

Type Consolidation

• Consolidated Role and Status types
• Reuse @Lumo types: ConversationPriv, ConversationPub, MessagePub
• Unified IncomingMessage and NormalizedMessage into MessageForStore
• Improved naming of OpenAI chat/responses message types and converter functions

Code quality

• Console shim suppresses upstream store log floods, maps to pino logger
• Added create-patch helper script
• Added initial unit tests for ConversationStore
• Use fake-indexeddb in mock mode instead of fallbackStore

Documentation

• Added Home Assistant setup guide (#20)
• Added heads up about official Lumo API in README and HA howto
• Updated docs on new ConversationStore and package structure

v0.4.0

Changelog

Breaking Changes

• Config renamed: cli.localActions.fileReads.maxFileSizeKB → cli.localActions.fileReads.maxFileSize

Features

• Initial OpenClaw support and OpenAI compatibility hardening #45, thanks @wranglerdriver!
  OpenAI edge-field parity, improved error semantics, OpenClaw tool-call reliability, and configurable api.bodyLimit.
• Metrics endpoint
  New /metrics endpoint with Prometheus-compatible stats on API requests, tool calls, and sync. Disabled by default. Grafana dashboard included.
• Commands wakeword
  Trigger commands without slash using a wakeword (e.g., "tamer title").
• Instructions injection
  Instructions now prepend to user messages (matching WebClient). New instructions.injectInto config for first/last message injection.
• Initial nanocoder support

Bug Fixes

• Token refresh on startup - Expired tokens now auto-refresh at startup (#37)
• Tool call validation and parsing - Fixed valid tool calls being marked invalid in non-streaming requests. Regular JSON blocks without a name field no longer treated as invalid tool calls.
• Log format - Fixed output format issues; now uses singleLine for pino-pretty (#38)

Refactoring

• Unified streaming/non-streaming handling in /v1/responses and /v1/chat/completions
• Simplified tool deduplication: removed call_id tracking from ConversationStore

Other

• Updated documentation for auth troubleshooting, wakeword usage, and API client compatibility (OpenClaw, nanocoder)
• Synced proton upstream files to 6dfd1f79

v0.3.0

Changelog

Breaking Changes

Commands & Binaries

• Consolidated all commands into tamer: tamer-auth → tamer auth, tamer-server → tamer server
• Docker: Renamed container from lumo-tamer-app to lumo-tamer
• Removed Makefile: Use npm commands instead

Configuration

• Instructions
  • instructions.default → instructions.fallback
  • Removed instructions.append: use instructions.template instead
  • Removed cli.instructions.default: use cli.instructions.template instead
• Server
  • tools.enabled → customTools.enabled
  • tools.enableWebSearch → enableWebSearch
• CLI
  • tools.enabled → localActions.enabled
  • Moved fileReads and executors under localActions

Conversations

Stateless requests (without conversation identifiers) are no longer synced to Lumo.

New Features

• Tool call support on /v1/chat/completions
• Handlebars-based instruction templates: Flexible instruction assembly with instructions.template
• Custom tool prefixes: Add customTools.prefix to namespace user tools (default: user:)

Improvements

• Improved custom tool instructions to reduce misrouted calls
• Early bounce for misrouted tool calls: detect when Lumo routes custom tools through its native pipeline and bounce them back immediately instead of waiting for the full response
• CLI executors passed dynamically to instructions
• Better conversation handling: stateless requests are truly stateless, fixing duplicate conversations and auto-title generation

Bug Fixes

• Fix conversation format mismatches with tool call outputs
• Fix invalid conversation continuations

Documentation

• Split tools documentation into custom-tools.md and local-actions.md
• Reworked README sections on Custom Tools, Local Actions, and API clients

Notes

This is an early release. Expect rough edges. Only tested on Linux. See the README for setup instructions.

v0.2.0

Changelog

Breaking Changes

Renamed config options

• tools.enableFileReads → fileReads.enabled
• conversation.sync fields: spaceName → projectName, spaceId → projectId
• conversations.deriveIdFromFirstMessage → conversations.deriveIdFromUser . Same goal, different behavior.

Features

• Handle Lumo's "confused" tool calls: bounce custom tools calls mistakenly routed through Lumo's native SSE tool pipeline so Lumo can try again via text messages. (#27)
• Add tool calls to /v1/responses (non-streaming) (#26)
• CLI: limit file reads to small, supported (non-binary) files only (#12)
• Add Lumo API mock for testing, based on Proton WebClients mock (#5)
• Add Vitest test suite: unit, integration (OpenAI endpoints), and e2e tests (#28)

Improvements

• /v1/responses and /v1/chat/completions: Comply to the OpenAI API streaming spec (#21)
• Be more verbose about Lumo errors: find & log error messages deep in HTTP response and JSON body
• Show autorefresh config in auth status
• Change user-facing "space" wording to "project"
• Add up-to-date doc on tools

Bug Fixes

• Fix not sending Lumo full Turns when sending tool call result
• Fix API returning empty response when raw JSON detector can't find JSON end (#25)
• Fix empty response on duplicate messages (#25)
• Fix auth-status crashing on uninitialized logger (#13)
• Clean up auth logger (#4)
• Always show warnings & errors in terminal, even when logging to a file (#18)
• Don't load existing conversations on startup (#6)
• Pass through Lumo API errors properly to logs and API client (c83b55b)

Notes

This is an early release. Expect rough edges. Only tested on Linux. See the README for setup instructions.

v0.1.0 - First public release

First public release of lumo-tamer.

Features

• OpenAI-compatible API (/v1/responses, /v1/chat/completions): use Lumo with any OpenAI-compatible client
• CLI: interactive mode with file read/create/edit support and command execution
• Conversation sync: encrypted conversation persistence, synced with Proton so you can continue on lumo.proton.me or mobile
• Tool support: experimental function calling via the API
• Multiple auth methods: browser-based, SRP login, or rclone token import
• Encrypted secret storage: session tokens stored in an AES-256-GCM vault, key in OS keychain or Docker secret
• Docker support: containerized setup with Docker Compose

Notes

This is an early release. Expect rough edges. Only tested on Linux. See the README for setup instructions.