Agents

Agent Status Model Preset Memory stores Tools Prompt sources Revision Action
Lumen
lumen
CURRENT gb10_1_openai_openai_compatible_deepseek_v4_1_flash_1017
agent_default
creative_brainstorming 1
lumen [identity] [rw]
177
current-agent authority
1 368 Loaded
Lumen
lumen_fast_uncensored
invalid
Model endpoint offline: nvidia_5090___qwen3.5-9b-uncensored-hauhaucs-aggressive
nvidia_5090___qwen3.5-9b-uncensored-hauhaucs-aggressive
agent_default
general_assistant 1
agent-lumen_fast_uncensored-1af60847 [identity] [rw]
122
current-agent authority
0 16
Uncensored
uncensored-agent
degraded
Optional source source_1: tool unavailable (get_identity)
gb10_1_vllm_qwen3_8_flash_next
agent_default
inherit 1
agent-uncensored-agent-f7aae8b4 [identity] [rw]
270
current-agent authority
1 29

Current Agent

Revision 368. All edits apply to the loaded current agent.

Memory Stores

Memory bindings are persisted with this agent in host-mounted /persist/config/agents.json (state/config/agents.json on the host). Changes to an existing agent are written immediately and also may be saved explicitly.

MCP Tools

Tools selected here belong to Lumen and are the authoritative capability policy for the agent-scoped MCP surface. After select_agent loads this agent as the Control Plane current agent, tools/list on port 11223 exposes these available tools plus the stable agent-interface tools. tools/call re-authorizes every call against that same current agent.

LLM Control Plane — Built-in Tools

Native capabilities provided by this Control Plane. They are grouped by logical subsystem and remain governed by the same agent AllowedTools policy.

42 tools
4 / 4 enabled for this agent
3 / 3 enabled for this agent
7 / 7 enabled for this agent
6 / 6 enabled for this agent
7 / 7 enabled for this agent
4 / 4 enabled for this agent
0 / 5 enabled for this agent
3 / 4 enabled for this agent
2 / 2 enabled for this agent

External MCP Server Tools

Federated tools advertised by configured downstream MCP servers. Each server remains visually separate from native LLM Control Plane capabilities.

312 tools
7 / 7 enabled for this agent
1 / 1 enabled for this agent
0 / 11 enabled for this agent
11 / 11 enabled for this agent
0 / 1 enabled for this agent
13 / 13 enabled for this agent
0 / 8 enabled for this agent
5 / 5 enabled for this agent
6 / 6 enabled for this agent
1 / 1 enabled for this agent
20 / 20 enabled for this agent
0 / 10 enabled for this agent
0 / 7 enabled for this agent
0 / 10 enabled for this agent
11 / 11 enabled for this agent
0 / 1 enabled for this agent
7 / 7 enabled for this agent
0 / 12 enabled for this agent
0 / 9 enabled for this agent
0 / 9 enabled for this agent
46 / 62 enabled for this agent
Available to agent Status Tool Timeout (s) Description
available storyboarder__accept_beat_proposal
blank = 600
Commit a beat proposal. Destructive: every existing beat on this storyboard and the artifacts attached to it are removed and replaced. Pass the beats exactly as generate_beats returned them. Metadata-only.
available storyboarder__accept_lore
blank = 600
Promote one or more proposed lore notes into canon in a single call. Pass a JSON array of lore note ids; a single note is an array of one. Canon notes are the only notes used as generation context and indexed for retrieval. Each item is reported independently; a note that is already canon is reported as already_canon rather than failing the batch. Metadata-only.
available storyboarder__accept_story_proposal
blank = 600
Commit a story proposal as the storyboard's script. Destructive: the existing story is replaced, not appended, so pass the exact proposal text that was reviewed. Metadata-only.
available storyboarder__add_beat
blank = 600
Append one or more beats to a storyboard in a single call. Pass a JSON array of beat specs; a single beat is an array of one. Each item is reported independently; a partial failure does not roll back the items that succeeded. Metadata-only.
available storyboarder__add_lore_subject
blank = 600
Add an operator-defined subject to a project's lore-subject catalog, so the universe can use subjects beyond the general starter set (e.g. 'Technology', 'Magic System'). The name is unique within the project. Metadata-only.
available storyboarder__create_backup
blank = 600
Write a consistent, unencrypted snapshot of the application database to the backup directory. Metadata-only.
available storyboarder__create_generator
blank = 600
Create a generator from an enumerated tool. Supply the endpoint key, the tool name, the stage it targets, the artifact role it produces, and where its prompt comes from. The parameter list is read from the tool's live schema; the values default to the schema's declared defaults until update_generator_params sets them.
available storyboarder__create_lore
blank = 600
Create lore notes from supplied text, verbatim. Unlike propose_lore, which asks the LLM to invent a note, this stores exactly the title and body the caller provides, with aliases and an explicit status. Pass a JSON array of note specs; a single note is an array of one. Each item is reported independently. A canon note is indexed for retrieval on creation; when the lore policy requires acceptance approval, a canon note is refused unless status is 'proposed' or auto-accept is enabled. Metadata-only.
available storyboarder__create_storyboard
blank = 600
Create a new storyboard in a project. Slug must be unique within the project. Metadata-only.
available storyboarder__delete_generator
blank = 600
Delete one or more generators in a single call. Pass a JSON array of generator keys; a single generator is an array of one. A generator still bound to a stage fails only its own item, so a stage is never left unbound by accident. Each item is reported independently. Metadata-only.
available storyboarder__delete_lore
blank = 600
Soft-delete one or more lore notes in a single call. Pass a JSON array of lore note ids; a single note is an array of one. The note keeps its row and is marked Deleted, excluded from generation and retrieval, and removed from the vector index; it can be restored. Deleting a note with children is refused unless the lore policy permits it. When the lore policy requires deletion approval, the call is refused unless auto-delete is enabled. Each item is reported independently. Metadata-only.
available storyboarder__delete_project
blank = 600
Delete one project and everything it owns: storyboards, beats, lore notes and aliases, and render jobs. This is a hard delete, not reversible from the API. Two separate guards apply. First, the caller must echo the project slug back in confirm_slug; a mismatch is refused so a wrong slug cannot delete a universe. Second, the lore policy's auto-delete-projects switch must be on, otherwise the call is refused with approval_required and the operator deletes it on the Projects page. Metadata-only.
available storyboarder__enumerate_endpoint_tools
blank = 600
Enumerate the tools an endpoint exposes, with each tool's schema-declared parameters. This is the first step of creating a generator: enumerate, pick a tool, then create_generator. The schema supplies the parameter list; the values are set locally.
available storyboarder__generate_beats
blank = 600
Ask the built-in beat generator to break this storyboard's story into beats, honouring the narration-style guidance, beat count, and beat length. Returns a PROPOSAL (each beat's narration, image_prompt, motion_prompt, intent, and planned seconds) and writes nothing; call accept_beat_proposal to commit it, which replaces every existing beat and its artifacts. Metadata-only.
available storyboarder__generate_story
blank = 600
Ask the built-in story writer to propose a story from the operator's request, using the project universe, canon lore, and this storyboard's story style note. Returns a PROPOSAL and does not touch the stored script; call accept_story_proposal to commit it, which replaces the current story wholesale. Metadata-only.
available storyboarder__generate_storyboard_music_prompt
blank = 600
Ask the configured LLM to write the audio bed prompt for this storyboard from its script and canon lore — the same synthesis the UI's Music tab runs, reached through the MCP surface. Returns a PROPOSAL and writes nothing; call set_storyboard_music_prompt with the returned text to commit it, or pass persist=true to commit it directly. Metadata-only.
available storyboarder__get_generator
blank = 600
Return one or more generators in full, including each schema-declared parameter list and the local values set for this pipeline. Pass a JSON array of generator keys; a single generator is an array of one. Each item is reported independently; an unknown key fails only its own item. Metadata-only.
available storyboarder__get_job_status
blank = 600
Return the status of one or more render jobs in a single call. Pass a JSON array of job ids; a single job is an array of one. This is the polling path for a batch of queued stages, so a caller can watch several jobs without a call per id. Each item is reported independently; an unknown id fails only its own item. Metadata-only.
available storyboarder__get_lore_note
blank = 600
Return one or more lore notes with full bodies and immediate children in a single call. Pass a JSON array of lore note ids; a single note is an array of one, which returns that note's fields directly. Each item is reported independently; a missing id fails only its own item, so one bad id does not discard the notes that resolved. Metadata-only.
available storyboarder__get_prompt_guide
blank = 600
Read the prompting guides for one or more generators: the operator's notes on how each specific model wants prompts written. Pass a JSON array of generator keys; a single generator is an array of one. Each item is reported independently; an unknown key fails only its own item. Metadata-only.
available storyboarder__get_service_info
blank = 600
Return the canonical runtime summary: wrapper URL, all four MCP endpoint URLs, transport revision, outputs base URL, runtime identity, and configured service inventory. Zero-argument; never produces artifacts.
available storyboarder__get_storyboard
blank = 600
Return the ordered beat list for one storyboard, including prompts and planned durations. Omit storyboard_slug to get the project's first storyboard. Metadata-only.
available storyboarder__get_vram_policy
blank = 600
Read the VRAM clearing policy: off, after each generation, or after a batch of N generations.
available storyboarder__import_artifact
blank = 600
Fetch an artifact server-to-server into this service's outputs tree, verify integrity, and commit atomically. destination_path is a logical path relative to the outputs root, e.g. 'image/psiloc/imported.png'. overwrite=false refuses an existing destination.
available storyboarder__list_artifacts
blank = 600
List recently produced artifacts newest-first with canonical served URLs. Metadata-only; never produces artifacts.
available storyboarder__list_available_samplers
blank = 600
Return the currently selectable MOSS samplers, queried live from the configured endpoint.
available storyboarder__list_available_voices
blank = 600
Return the currently selectable Kokoro voices, queried live from the configured endpoint.
available storyboarder__list_available_weight_files
blank = 600
Return the currently selectable MOSS weight files, queried live from the configured endpoint. Never exposes host paths.
available storyboarder__list_backups
blank = 600
List the database backups on disk, newest first. Metadata-only.
available storyboarder__list_endpoints
blank = 600
List the configured service endpoints with their keys, URLs, and enabled state. Use the key with enumerate_endpoint_tools and create_generator.
available storyboarder__list_generators
blank = 600
List the configured generators with their keys, types, stages, endpoints, tool names, and enabled state. Use a key with get_generator, update_generator, test_generator, or render_stage's tool_key.
available storyboarder__list_lore
blank = 600
List lore notes for one project as a flat set with parent links, so a client can rebuild the tree. Filter by status ('canon' or 'proposed'); omit to return both. Metadata-only.
available storyboarder__list_lore_subjects
blank = 600
List the project's controlled lore-subject catalog: the valid subject names a note can be tagged with, plus each subject's guidance. The catalog is per-project and operator-extensible, so an agent can enumerate exactly which subjects exist before tagging. Metadata-only.
available storyboarder__list_projects
blank = 600
List configured project profiles with their storyboard and canon-lore counts. Metadata-only.
available storyboarder__list_stage_bindings
blank = 600
List which generator is bound to each render stage, including any configured fallback.
available storyboarder__list_storyboards
blank = 600
List the storyboards belonging to one project, with beat counts. Metadata-only.
available storyboarder__propose_lore
blank = 600
Ask the configured LLM endpoint to write one new lore note from a user prompt, using only canon lore as context. The result is stored as a proposed note and is NOT canon until accept_lore is called. Metadata-only.
available storyboarder__purge_orphan_artifacts
blank = 600
Find and optionally delete files on the outputs volume that no artifact record references. A file is referenced only when its exact relative path belongs to an artifact row, so files left by abandoned renders, deleted records, or removed projects are orphans. Referenced files are never touched. With dry_run=true (the default) nothing is deleted and the orphans are returned; call again with dry_run=false to delete them. Metadata-only.
available storyboarder__refresh_generator
blank = 600
Re-read the generator's tool schema from its endpoint and update the declared parameter list. Never overwrites the local values; a value whose parameter disappeared is reported as orphaned.
available storyboarder__reject_lore
blank = 600
Discard one or more proposed lore notes in a single call. Pass a JSON array of lore note ids; a single note is an array of one. Each item is reported independently; a canon note or a note with children fails only its own item. Rejection is not recursive. Metadata-only.
available storyboarder__remove_beat
blank = 600
Delete one or more beats in a single call. Pass a JSON array of beat ids; a single beat is an array of one. Each item is reported independently; a beat that does not exist fails only its own item. Artifacts attached to removed beats are removed with them. Metadata-only.
available storyboarder__render_stage
blank = 600
Run one render stage for one storyboard. Omit storyboard_slug to render the project's first storyboard. Omit tool_key to use the generator bound to the stage; supply tool_key to run a specific generator for this call only, without changing the binding. With wait=true (default) the call blocks until the stage finishes and returns the produced artifacts. With wait=false it returns the job id immediately; poll get_job_status and then list_artifacts. variant='test' turns the run into its preview form: for 'animate' it pins a random seed and a reduced frame so the clip lands at beat_{NN}_video_test.mp4 under role beat_clip_test; for 'beat_assemble' it muxes that test clip to beat_assembled_test; for 'assemble' it concatenates the beat_assembled_test deliverables into final_test.mp4 under role final_mux_test. Preview output never replaces production output.
available storyboarder__restore_backup
blank = 600
Restore a backup. mode 'merge' (default) inserts missing rows and overwrites rows the backup has a newer UpdatedAt for; mode 'overwrite' replaces the live database wholesale. A pre-restore snapshot is always taken first. Metadata-only.
available storyboarder__restore_lore
blank = 600
Restore one or more soft-deleted lore notes in a single call. Pass a JSON array of lore note ids; a single note is an array of one. A restored note returns to Proposed, not Canon, so it must be accepted again before it influences generation; this keeps a restored note from silently re-entering canon. Each item is reported independently. Metadata-only.
available storyboarder__rewrite_beat_field
blank = 600
Ask the configured LLM to rewrite one beat field (narration, image_prompt, or motion_prompt) using the storyboard script, the beat's stated intent, and neighboring beats as read-only context. Returns a PROPOSAL; it does not change the beat. Call update_beat_field to accept it. Metadata-only.
available storyboarder__set_generator_negative_prompt
blank = 600
Set or clear a generator's negative prompt. This is generator-owned, not per beat: every artifact the generator produces uses it. Pass an empty value to clear it. Metadata-only.
available storyboarder__set_lore_aliases
blank = 600
Replace the alias set for one lore note. Aliases are the surface forms that refer to the note in prose; they are what the Story tab highlights and what the story-side lore append matches against. The note keeps its own definition; aliases never hold one. A surface must be unique within the note's project. Canon notes' aliases are the only ones used for highlighting and generation. Metadata-only.
available storyboarder__set_lore_tags
blank = 600
Replace the subject tags on one lore note. Pass a JSON array of {subject, policy} where subject is a catalog name and policy is one of Always, MentionGated, PullOnly, or StageScoped. This is what drives retrieval: Always is injected for every visual consumer, MentionGated only when the text mentions it, PullOnly never automatically, StageScoped when the consumer is that stage. Metadata-only.
available storyboarder__set_prompt_guide
blank = 600
Set or clear the prompting guide for one generator. Pass an empty guide to clear it. Metadata-only.
available storyboarder__set_stage_binding
blank = 600
Bind a generator to a render stage, replacing any existing binding for that stage. Optionally set a fallback generator used when the primary is disabled.
available storyboarder__set_storyboard_beat_generation
blank = 600
Set the storyboard's beat-generation settings: beat_count, beat_seconds, and narration_words_per_second. These are what a later generate_beats uses when no per-call override is supplied, and they decide the spoken word budget. Set them BEFORE generating: leaving a GUI default in place while generating with different call arguments sizes the arc and the narration budget from different numbers. The duration is clamped to the pipeline's ceiling and the count must be positive. Metadata-only.
available storyboarder__set_storyboard_music_prompt
blank = 600
Set the audio bed prompt for this storyboard: mood, instrumentation, tempo. Storyboard scope, not project scope. Metadata-only.
available storyboarder__set_storyboard_script
blank = 600
Set the script or backstory this storyboard dramatizes. This is the source of truth the rewrite tools read as context. Metadata-only.
available storyboarder__set_storyboard_writing_guidance
blank = 600
Set optional operator guidance on how this telling is WRITTEN. story_style is advice on the story prose; narration_style is applied on top of the built-in spoken-clarity rules in both beat generation and narration rewrite. Blank or omitted leaves the field unset, which means the built-in defaults apply. Metadata-only.
available storyboarder__set_vram_policy
blank = 600
Set the VRAM clearing policy. mode is one of: off, after_each_generation, after_batch. batch_size is required for after_batch and ignored otherwise.
available storyboarder__test_generator
blank = 600
Run one cheap probe render against the generator's tool to learn where it puts its artifact URL, then record that path. Costs one real generation at the smallest settings the schema allows. Supply a short probe prompt; there is no universal default.
available storyboarder__update_beat_field
blank = 600
Write one beat field directly. This is how a rewrite proposal is accepted. Metadata-only.
available storyboarder__update_beat_metadata
blank = 600
Change a beat's structural fields: order, kind, planned seconds, or lore reference. Omit a parameter to leave it unchanged. Metadata-only.
available storyboarder__update_generator
blank = 600
Update a generator's identity and routing fields, including which tool call it uses. Changing mcp_tool_name re-reads that tool's schema from the endpoint and replaces the parameter list; the artifact path is cleared because it was learned from a different tool. Does not touch parameter values; use update_generator_params for those.
available storyboarder__update_generator_params
blank = 600
Set the local values for a generator's parameters. Supply a JSON object mapping parameter names to values, e.g. {"width":768,"height":1344}. Names must exist in the generator's schema-declared parameter list; unknown names are rejected rather than silently stored.
available storyboarder__update_lore
blank = 600
Revise an existing lore note's title, body, kind, parent, or order. Omitted fields are left unchanged. Editing a canon note re-embeds it so retrieval reflects the new text. Aliases are not changed here; use set_lore_aliases. Metadata-only.
available storyboarder__vision_analyze
blank = 600
Analyze an image with the configured vision model and return a textual description. Supply either an image_url (a URL the vision service fetches itself) or an artifact_id already in the store; the artifact's canonical URL is used. Use this to inspect a rendered frame or reference image. Never produces artifacts.
0 / 9 enabled for this agent
0 / 9 enabled for this agent
0 / 9 enabled for this agent
0 / 9 enabled for this agent
0 / 9 enabled for this agent
0 / 9 enabled for this agent
0 / 5 enabled for this agent
9 / 9 enabled for this agent
0 / 6 enabled for this agent
0 / 5 enabled for this agent
4 / 4 enabled for this agent
0 / 7 enabled for this agent

Skills

Authoritative skill enablement for Lumen. Enabling a skill here is content permission only: the agent must also have the skills_* tools to discover and load it. Groups are defined on Data → Skills.

10 / 10 enabled

Ungrouped skills

10 / 10 enabled for this agent
Enabled for agent Skill Description
ace-step-text-to-music
local
Generate a song with the ACE-Step 1.5 audio service from an explicit style prompt plus lyrics, then report the persisted artifact and attachment receipt.
control-plane-write-skills
local
Create, revise, organize, validate, and publish skills for the LLM control plane. Use when asked to write a new skill, turn a repeated workflow into a skill, repair an existing skill package, organize a skill collection, or import finished skills through the control plane MCP endpoint.
csharp-coding
local
Write, review, and refactor C# code following current .NET conventions. Use when creating or editing .cs files, implementing .NET console or library code, fixing compiler or analyzer warnings, or reviewing C# projects for structure and style consistency.
flux2-text-to-image-workflow
local
Generate an image with the Flux 2 Klein service and document the observed process.
json-schema-helper
local
Author, validate, and troubleshoot JSON Schema documents. Use when asked to write a JSON Schema for a payload or config file, validate JSON data against a schema, explain why a document fails validation, or choose the right draft version and keywords for a schema.
markdown-formatter
local
Normalize and lint Markdown documents for consistent structure, whitespace and heading hierarchy. Use when asked to format, tidy, clean up, or lint a Markdown file, README, changelog, or notes document; also use before publishing docs to catch broken heading levels, inconsistent list markers, trailing whitespace and malformed tables.
pdf-summarizer
local
Summarize a PDF attachment supplied in the conversation, using only text the user or client actually provides. Use when asked to summarize, digest, outline, or extract key points from a PDF; also use when a PDF arrives as an attachment reference, a vault path, or a URL and the agent must decide whether it can be read at all before promising a summary.
storyboard-creation
local
Build a storyboard end to end in Storyboarder over MCP, driven by the harness's internal LLM rather than the calling agent's own prose. Use when turning a premise or story idea into a rendered storyboard, adding a telling to an existing universe, or writing contiguous beats with image and motion prompts that stay consistent with canon. Contains a definition of done, a cost order, and the specific failure modes that have caused incomplete or overpriced runs.
storyboard-lore
local
Read, audit, and extend Storyboarder canon lore over MCP without corrupting it. Use when asked to review lore for contradictions or ambiguity, to check what a downstream agent will read literally, to add a canon note that fits an existing universe, or to explain why a storyboard beat cites something canon never said. Includes a definition of done so an audit is finished before it is reported.
toml-formatter
local
Normalize and lint TOML configuration files for consistent structure, key ordering, spacing and quoting. Use when asked to format, tidy, clean up, or lint a TOML file such as Cargo.toml, pyproject.toml, a config file, or any .toml document; also use before committing config changes to catch duplicate keys, invalid table headers and inconsistent formatting.

Boot / Prompt Sources

MCP source outputs are composed into the system layer. Saved sources are persisted with this agent in host-mounted /persist/config/agents.json (state/config/agents.json on the host). This section is the authority for which prompt-source tools run and whether a source is required.

Tool Arguments

Generated from the selected MCP tool's inputSchema. Enum and boolean choices are populated automatically.

Advanced raw arguments JSON

Tool Arguments

Generated from the selected MCP tool's inputSchema. Enum and boolean choices are populated automatically.

Advanced raw arguments JSON