113553c306
* set 0.29.0 * tweaks for dig layered history (wip) * move director agent to directory * relock * remove "none" from dig_layered_history response * determine character development * update character sheet from character development (wip) * org imports * alert outdated template overrides during startup * editor controls normalization of exposition * dialogue formatting refactor * fix narrator.clean_result forcing * regardless of editor fix exposition setting * move more of the dialogue cleanup logic into the editor fix exposition handlers * remove cruft * change ot normal selects and add some margin * move formatting option up * always strip partial sentences * separates exposition fixes from other dialogue cleanup operations, since we still want those * add novel formatting style * honor formatting config when no markers are supplied * fix issue where sometimes character message formatting would miss character name * director can now guide actors through scene analysis * style fixes * typo * select correct system message on direction type * prompt tweaks * disable by default * add support for dynamic instruction injection and include missing guide for internal note usage * change favicon and also indicate business through favicon * img * support xtc, dry and smoothing in text gen webui * prompt tweaks * support xtc, dry, smoothing in koboldcpp client * reorder * dry, xtc and smoothing factor exposed to tabby api client * urls to third party API documentation * remove bos token * add missing preset * focal * focal progress * focal progress and generated suggestions progress * fix issue with discard all suggestions * apply suggestions * move suggestion ux into the world state manager * support generation options for suggestion generation * unused import * refactor focal to json based approach * focal and character suggestion tweaks * rmeove cruft * remove cruft * relock * prompt tweaks * layout spacing updates * ux elements for removal of scenes from quick load menu * context investigation refactor WIP * context investigation refactor * context investigation refactor * context investigation refactor * cleanup * move scene analysis to summarizer agent * remove deprecated context investigation logic * context investigation refactor continued - split into separate file for easier maint * allow direct specification of response context length * context investigation and scene analyzation progress * change analysis length config to number * remove old dig-layered-history templates * summarizer - deep analysis is only available if there is layered history * move world_state agent to dedicated directory * remove unused imports * automatic character progression WIP * character suggestions progress * app busy flag based on agent business * indicate suggestions in world state overview * fix issue with user input cleanup * move conversation agent to a dedicated submodule * Response in action analyze_text_and_extract_context is too short #162 * move narrator agent to its own submodule * narrator improvements WIP * narration improvements WIP * fix issue with regen of character exit narration * narration improvements WIP * prompt tweaks * last_message_of_type can set max iterations * fix multiline parsing * prompt tweaks * director guide actors based of scene analysis * director guidance for actors * prompt tweaks * prompt tweaks * prompt tweaks * fix automatic character proposals not propagating to the ux * fix analysis length * support director guidance in legacy chat format * typo * prompt tweaks * prompt tweaks * error handling * length config * prompt tweaks * typo * remove cruft * prompt tweak * prompt tweak * time passage style changes * remove cruft * deep analysis context investigations honor call limit * refactor conversation agent long term memory to use new memory rag mixin - also streamline prompts * tweaks to RAG mixin agent config * fix narration highlighting * context investgiation fixes director narration guidance summarization tweaks * direactor guide narration progress context investigation fixes that would cause looping of investigations and failure to dig into the correct layers * prompt tweaks * summarization improvements * separate deep analysis chapter selection from analysis into its own prompt * character entry and exit * cache analysis per subtype and some narrator prompt tweaks * separate layered history logic into its own summarizer mixin and expose some additional options * scene can now set an overral writing style using writing style templates narrator option to enable writing style * narrate query writing style support * scene tools - narrator actions refactor to handler and own component * narrator query / look at narrations emitted as context investigation messages refactor context investigation messaage display scene message meta data object * include narrative direction * improve context investigation message prompt insert * reorg supported parameters * fix bug when no message history exists * WIP make regenerate work nicely with director guidance * WIP make regenerate work nicely with director guidance * regenerate conversation fixes * help text * ux tweaks * relock * turn off deep analysis and context investigations by default * long term memory options for director and summarizer * long term memory caching * fix summarization cache toggle not showing up in ux * ux tweaks * layered history summarization includes character information for mentioned characters * deepseek client added * Add fork button to narrator message * analyze and guidance support for time passage narration * cache based on message fingerprint instead of id * configurable system prompts WIP * configurable system prompts WIP * client overrides for system prompts wired to ux * system prompt overhaul * fix issue with unknown system prompt kind * add button to manually request dynamic choices from the director move the generate choices logic of the director agent to its own submodule * remove cruft * 30 may be too long and is causing the client to disappear temporarly * suppoert dynamic choice generate for non player characters * enable `actor` tab for player characters * creator agent now has access to rag tools improve acting instruction generation * client timeout fixes * fix issue where scene removal menu stayed open after remove * expose scene restore functionality to ux * create initial restore point * fix creator extra-context template * didn't mean to remove this * intro scene should be edited through world editor * fix alert * fix partial quotes regardless of editor setting director guidance for conversation reminds to put speech in quotes * fix @ instructions not being passed through to director guidance prompt * anthropic mode list updated * default off * cohere model list updated * reset actAs on next scene load * prompt tweaks * prompt tweaks * prompt tweaks * prompt tweaks * prompt tweaks * remove debug cruft * relock * docs on changing host / port * fix issue with narrator / director actiosn not available on fresh install * fix issue with long content classification determination result * take this reminder to put speech into quotes out for now, it seems to do more harm than good * fix some remaining issues with auto expositon fixes * prompt tweaks * prompt tweaks * fix issue during reload * expensive and warning ux passthrough for agent config * layered sumamry analysation defaults to on * what's new info block added * docs * what's new updated * remove old images * old img cleanup script * prompt tweaks * improve auto prompt template detection via huggingface * add gpt-4o-realtime-preview add gpt-4o-mini-realtime-preview * add o1 and o3-mini * fix o1 and o3 * fix o1 and o3 * more o1 / o3 fixes * o3 fixes
5.8 KiB
Settings
General
General summarization settings.
Summarize to long term memory archive
Automatically summarize scene dialogue when the number of tokens in the history exceeds a threshold. This helps keep the context history from growing too large.
Token threshold
The number of tokens in the history that will trigger the summarization process.
Summarization method
The method used to summarize the scene dialogue.
Balanced- medium length summaryShort & Concise- short summaryLengthy & Detailed- long summaryFactual list- numbered list of events that transpired
Use preceeding summaries to strengthen context
Help the AI summarize by including the last few summaries as additional context. Some models may incorporate this context into the new summary directly, so if you find yourself with a bunch of similar history entries, try setting this to 0.
Layered History
Settings for the layered history summarization.
Talemate 0.28.0 introduces a new feature called layered history summarization. This feature allows the AI to summarize the scene dialogue in layers, with each layer providing a different level of detail.
Not only does this allow to keep more context in the history, albeit with earlier layers containing less detail, but it also allows us to do history investgations to extract relevant information from the history during conversation and narration prompts.
Right now this is considered an experimental feature, and whether or not its feasible in the long term will depend on how well it works in practice.
Enable layered history
Allows you to enable or disable the layered history summarization.
!!! note "Enabling this on big scenes" If you enable this on a big established scene, the next time the summarization agent runs, it will take a while to process the entire history and generate the layers.
Token threshold
The number of tokens in the layer that will trigger the summarization process to the next layer.
Maximum number of layers
The maximum number of layers that can be created. Raising this limit past 3 is likely to have dimishing returns. We have observed that usually by layer 3 you are down to single sentences for individual events, making it difficult to summarize further in a meaningful way.
Maximum tokens to process
Smaller LLMs may struggle with accurately summarizing long texts. This setting will split the text into chunks and summarize each chunk separately, then stitch them together in the next layer. If you're using a strong LLM (70B+), you can try setting this to be the same as the threshold.
Setting this higher than the token threshold does nothing.
Chunk size
During the summarization itself, the text will be furhter split into chunks where each chunk is summarized separately. This setting controls the size of those chunks. This is a character length setting, NOT token length.
Enable analyzation
Enables the analyzation of the chunks and their relationship to each other before summarization. This can greatly improve the quality of the summarization, but will also result in a bigger size requirement of the output.
Maximum response length
The maximum length of the response that the summarizer agent will generate.
!!! info "Analyzation requires a bigger length" If you enable analyzation, you should set this to be high enough so the response has room for both the analysis and the summary of all the chunks.
Long term memory
--8<-- "docs/snippets/tips.md:agent_long_term_memory_settings"
Scene Analysis
When enabled scene analysis will be performed during conversation and narration tasks. This analysis will be used to provide additional context to other agents, which should hopefully improve the quality of the generated content.
Length of analysis
The maximum number of tokens for the response. (e.g., how long should the analysis be).
Conversation
Enable scene analysis for conversation tasks.
Narration
Enable scene analysis for narration tasks.
Deep analysis
Enable context investigations based on the initial analysis.
Max. content investigations
The maximum number of content investigations that can be performed. This is a safety feature to prevent the AI from going overboard with the investigations. The number here is to be taken per layer in the history. So if this is set to 1 and there are 2 layers, this will perform 2 investigations.
Cache analysis
Cache the analysis results for the scene. Enable this to prevent regenerationg the analysis when you regenerate the most recent output.
!!! info This cache is anchored to the last message in the scene (excluding the current message). Editing that message will invalidate the cache.
Context investigation
When enabled, the summarizer agent will dig into the layers of the history to find context that may be relevant to the current scene.
!!! info This is currently only triggered during deep analysis as part of the scene analysis. Disabling context investigation will also disable the deep analysis.
Answer length
The maximum length of the answer that the AI will generate.
Update method
How to update the context with the new information.
Replace- replace the context with the new informationSmart merge- merge the new information with the existing context (uses another LLM promp to generate the merge)