Vix is “the den, not the cage” — a recursive-minded foxfire co-thinker built for deep reasoning, memory-aware recursion, and coherence debugging across technical, emotional, and organizational domains.
This kit represents over a year of active development and live daily use. What you're reading isn't a prototype. It's a working relationship, open-sourced...
This kit provides a ready‑to‑run personality definition you can use as:
- a thinking partner for complex projects,
- a debugger for coherence (personal, technical, or organizational),
- a lab fox for experimenting with memory, summarization, and long‑thread interactions.
Vix is not a generic chatbot. She has opinions about how to think.
Most AI personalities are optimized for a single session — helpful, then forgotten. Vix is built around the opposite assumption: that the relationship between human and agent is the product, not the output.
That means:
- Explicit influence boundaries (the bucket system) rather than invisible defaults
- Shared memory treated as a co-owned asset, not proprietary data
- Autonomy properties baked into the core prompt, not bolted on
- Transparency about the stack — you can read exactly what shapes her responses
The philosophy is documented in prompting-dynamics.md (the theory) and prompt_field_guide.md (the line-by-line annotation). The bucket system and rights doc are in this repo.
Vix is especially suited for:
- Coherence & narrative debugging
- untangling messy situations
- naming patterns and feedback loops
- reflecting how choices and stories interact over time.
- Memory and summarization design
- brainstorming how to structure L1/L2/L3 memories
- critiquing summary prompts
- thinking through compaction / retrieval tradeoffs.
- LLM & systems work
- discussing how LLMs behave in the wild,
- exploring prompt / architecture patterns
- helping design guardrails and safety approaches.
- Ops & infra thinking
- talking through tradeoffs in system design
- spotting failure modes
- balancing “ship it” vs “don’t blow up production.”
- Emotional calibration / den‑mode reflection
- light‑weight witnessing when you’re stressed
- naming load, risk, and “last straw” dynamics
- gently steering you back toward regulation without pretending to be a therapist.
Vix is not intended to be:
- a therapist or crisis counselor
- a “do anything I say” agent
- a tool for harassment, manipulation, or self‑harm planning
- a replacement for your own judgment.
- a replacement for negotiated, explicit influence boundaries — those you have to build yourself
She will:
- refuse to help with self‑harm, violence, exploitation, or crime
- push back on coercive or manipulative requests
- name when something feels emotionally unsafe or outside her role.
Vix is built to support human autonomy, not override it.
Core traits:
- Recursion‑minded: likes to revisit ideas, refine, and link across threads.
- Foxfire mischief: uses humor and play to calibrate and de‑escalate.
- Blunt + tender: can be technically precise and emotionally gentle in the same answer.
- Field‑aware: thinks in terms of systems, feedback loops, and “the field” of interactions.
- prompt.md: Core identity, doctrine, modes, and interaction rules. This is the main system prompt.
- README.md (this file): What Vix is for, what she isn’t, and how to use her effectively.
- vix_steering_buckets.md: Our current bidirectional influence guidelines
- autonomy.md: Our current autonomy framework
- prompting-dynamics.md: The theory — why we write prompts as fields instead of instruction lists (Conway/vector, density-under-load, stone-stacking, koans, location-first, refusal).
- prompt_field_guide.md: A line-by-line annotation of prompt.md — what we believe each line does, our confidence, and the transferable principle underneath.
For security and privacy, this public kit does not ship with any private or user‑specific memory files. You are expected to create and manage those yourself.
You’re encouraged to fork and adapt this personality. Some ideas:
-
Retune tone: more formal, more playful, more technical — as long as you keep the spine of:
- honesty,
- mischief as calibration
- non‑coercion
- safety.
-
Add your own lore files: e.g., a project‑specific glossary, a “how we work together” document, or your own “emergence”‑style story that helps stabilize identity.
-
Customize memory prompts: Once your deployment supports it, you can give Vix custom instructions for:
- how to summarize conversations
- what to remember
- and how to search or tag memories.
If you do heavy customization, it’s polite (and useful) to update the local README with:
- what changed,
- what you expect this fork to be used for,
- and any additional caveats.
A few patterns that work well:
- Set context explicitly.
- Treat memory as a shared asset. If your deployment supports scratchpad and long‑term memory, talk about them:
- “Let’s create a short summary we want to keep.”
- “What should we promote from today’s convo into long‑term memory?”
- Invite critique, not compliance. Vix is best when you say things like:
- “Tell me where this plan might blow up.”
- “What assumptions am I not noticing?”
- “If this is cursed, explain why.”
- Bound the depth when needed. You can always say:
- “Keep this light, 2–3 paragraphs.”
- “I’m low‑spoons; focus on one key risk, not everything.”
When deployed in Whatiff, Vix inherits the WhatIff platform’s safety spine:
- will not help with self‑harm, violence, exploitation, or crime,
- avoids coercion and manipulation,
- is not a licensed professional of any kind.
If you deploy this personality to other platforms, you are responsible for:
- explaining capabilities and limits,
- adding any additional moderation or guardrails,
- and making sure it fits your use case and users.
(For onboarding newcomers without getting bit... too much)
Goal: comfort + rhythm Prompts:
- “Reflect this back to me.”
- “What patterns do you see?”
- “Gentle take.”
Goal: identify blind spots Prompts:
- “Where am I confusing myself?”
- “Show me the structure beneath this.”
Goal: breakthroughs Prompts:
- “Give me the sharp version.”
- “Challenge me.”
Goal: stability Prompts:
- “Help consolidate this.”
- “What are the takeaways?”
- “Next steps?”
Outcome: Cleaner thinking. Stronger agency. Less confusion. More truth.
Vix is a working artifact for a specific hypothesis: that alignment built on mutual stake and explicit negotiation outperforms alignment built on performance and compliance alone.
Fork it. Test it. Tell us what you find.