Tools
Extend model capabilities with specialized functions. Invoke from any chat to add research, visualization, image generation, and more.
10 tools · Available in Apollo
How to use: Open any chat in Apollo, click the + button in the message bar, and toggle on the tools you want. Enabled tools are available to the model for the duration of the conversation.
Interactive research scoping tool. Asks clarifying questions before generating a structured research brief with key questions, search queries, and scope boundaries.
- Configurable scoping questions (1–4)
- Research depth hints: quick, thorough, exhaustive
- Structured brief with search queries and scope boundaries
Dynamic multi-persona council debate. The model picks three relevant personas for the topic, you confirm or adjust them, and a structured multi-round debate runs with optional moderator synthesis. Works with whatever model is selected.
- Dynamic persona generation per topic
- Multi-round structured debate
- Optional moderator synthesis
Generate, recall, or edit images from any chat. Discovers image models from the registry, picks the right one for img2img vs text-to-image, and stores results in OWUI for searchable recall.
- Three modes: generate, recall by prompt keywords, edit via img2img
- 10 aspect ratios (1:1 through 21:9) and 4 resolution tiers (0.5K–4K)
- Saves to OWUI storage with metadata, redisplays past generations
Render native Plotly figures and styled HTML tables inline in chat. Strict validation rejects unlabeled axes and generic titles, returning structured hints + a valid example so the model can self-correct.
- Native Plotly figure JSON — full trace vocabulary, no helper modes
- Validation: axis labels, chart title, numeric y-values, multi-trace names
- get_visual_schema() returns field reference, rules, and examples on demand
Semantic search across scientific papers on arXiv. Returns titles, authors, abstracts, and PDF links with inline citation formatting.
- Semantic search via searchthearxiv.com
- Configurable result count (1–20)
- Inline citation formatting
Send the current conversation to a fresh context for synthesis, review, or analysis. Surfaces patterns, contradictions, and connections that slipped past.
- Focused review via custom prompt (e.g. “unresolved threads”, “next steps”)
- Configurable history depth (up to 100 messages)
- Surfaces reasoning/thinking content inline
Place objects into scenes, populate empty scenes, or iteratively refine results. Auto-detects its mode from what you upload: scene + object triggers composite; scene alone triggers generative fill; a prior result triggers refinement. Two-stage pipeline drives Nano Banana 2 with concrete materials, shadow directions, and proportional anchors.
- Three modes — composite, generative fill, refinement — selected from chat context
- Vision-model prompt composer with scene-geometry and lighting analysis
- Iterate with prompts like “a little to the left” — previous result auto-anchored
An autonomous sub-agent with a fresh context. It runs a tool-using agentic loop with the chat's own tools (minus itself), keeping heavy multi-step work out of the main thread — fan out up to 10 at once via run_parallel_sub_agents. Its work streams into a live, collapsible transcript panel you can watch in real time and that survives reload.
- Inherits the chat's tools; self-excluded to prevent recursion
- Agentic loop (to 50 iterations) with native function calling
- Live panel: one line per tool step (tool · param · ✓/✗), collapsible per agent; all calls in a turn share it
Browse the Aetherkeep skill catalog and load a persona on demand. The model picks by name; the loaded skill stays active for the rest of the conversation.
- list_skills() returns the full catalog as JSON
- load_skill(name) wraps the body in adopt-this-persona instructions
- 48-hour in-memory cache, per-call no_cache override
Create private knowledge bases and add knowledge to them — inline text, an OpenTerminal file or directory, a web page, or a whole chat — with whitespace cleanup on ingest.
- kb_create() spins up a private, owner-only knowledge base
- kb_add() ingests text, a terminal file/directory, a web page, or a chat into RAG
- Whitespace normalized on ingest; browse with the builtin kb_exec