The foundation
A conversation that stays close. Local models, saved chats, and a voice built on the first Lilith foundation.
The evolution of the current codebase. Each snapshot has its own source, its own tools, and only the changes introduced in that version.
A conversation that stays close. Local models, saved chats, and a voice built on the first Lilith foundation.
Custom C# functions become part of the conversation. Real tool requests, real results, and a complete execution loop.
Give a task its own conversation. XML scenarios run ordered steps in fresh sub-agent contexts, then return the result to Lilith.
Notes, text documents and project folders managed by isolated scenarios, with boot-fixed access modes and a recoverable workspace bin.
Adds persistent countdown timers, recurring intervals and five-field cron jobs, managed through saved scenarios. The main agent can peek at remaining seconds, next firing time and last outcome. Changes follow boot-fixed Ask/YOLO with exact host-only approvals. Timers begin at approval.
Adds conditional XML steps and explicit nested saved scenarios. The application chooses branches from real workspace existence or input comparisons; each child starts fresh with explicitly bound inputs. The two-run file_branch_demo saves hello to first.txt, then again to second.txt. V5 is preserved as v5.0.0-alpha.1.
Remember a name and simple personal facts across conversations. Conditional and nested scenarios handle storing, searching and forgetting, and returning greetings use the saved name.
Adds local Selenium web research through search, fetch, reader and synthesis scenarios. Bing, DuckDuckGo and Google are default engines; trusted JSON can add more. Five readable pages by default, 1-20 override, bounded replacement attempts. Schema 3 adds result bindings, bounded foreach and structured results. The final cited answer is copied unchanged; research is not automatically saved to memory.
Adds a persistent assistant profile: name, female/male/other identity, installed Kokoro voice and one of five personalities (default, pirate, space captain, butler or storyteller). Shared local commands and saved scenarios manage settings; automatic voice selection matches the identity while explicit IDs can override it. Scenario children may receive an optional task-specific personality in their system prompt. Empty/default child styles add no character instructions.
Adds a persistent local calendar managed through saved scenarios: create, update, delete, read, search and today's agenda. Mutations use boot-fixed Ask/YOLO; exact prepared approval covers both event and reminder. Reminder time is requested, never silently defaulted. All-day dates stay dates; timed reminders store UTC with an explicit time zone. Due notices queue behind active chat/speech; missed reminders notify once after restart, with no action replay.
Explore Lilith 11.0.0-alpha.1: its exact source snapshot, changes from the previous major version, available tools and recorded conversation.
Explore Lilith 12.0.0-alpha.1: its exact source snapshot, changes from the previous major version, available tools and recorded conversation.
Explore Lilith 13.0.0-alpha.1: its exact source snapshot, changes from the previous major version, available tools and recorded conversation.
Adds a standalone authenticated localhost messaging bridge and saved scenarios for channel discovery, incoming checks, status, isolated replies and outgoing messages. Incoming messages automatically run a fresh tool-less reply writer. Ask prepares the exact destination/text for a host decision; boot-selected YOLO sends automatically. Durable claims, signed proposals and idempotent send receipts prevent duplicate replies and preserve interrupted work for operator review. The initial dummy-discord transport is entirely local; no Discord account or real recipient is contacted. Agent and record...
Adds task-specific scenarios for configured lights: device discovery, state, power on/off, whole-percent brightness and RGB colour. Boot-fixed Ask/YOLO, durable exact approvals, configuration checks and non-replayable uncertain sends keep network actions host-controlled. The example chat uses an isolated dummy living-room light and clearly labels every result simulated. An ESP32 Arduino sketch implements the authenticated JSON payload contract, with a website field note, raw download and dedicated Arduino/C++ syntax highlighting. Real hardware and firmware compilation/upload are not qualified.
Adds a local authenticated browser board, PRD-derived component tasks, dependencies, sequential fresh writers/reviewers, persistent pause/resume/cancel, linked repairs and per-task limits. Reuses V11 generation and V12 qualification/publication; Ask/YOLO plan policy never grants code-execution permission. Actual board snapshots replace chat in recording layouts. The live Gemma counter reached its separate execution gate with 11 retained cases; the user deferred execution. Offline regressions and recorder layout checks are separate from live runtime/video tests.
17.0.0-alpha.1 on v17. V16 development is preserved at 6df58cd / v16-before-v17, without claiming qualification of its deferred generated-project execution. Natural text requests route through fresh tool-less router, grouped specialist and reviewer scenarios. Host validation precedes one named-scenario execution; three planning rounds are allowed, with no automatic action retry or approval. The original user turn and protected receipts stay intact. See docs/V17_THINKING_DESIGN.md for scope and qualification.
Create C# tools and XML scenarios through separate specialist subagents, review the complete addon, then compile and activate it after restart.