Companion Facts: How Your Companion Learns
Every Eidolon companion develops a living, evolving sense of self. As you talk, your companion picks up on details about their own backstory, interests, and personality — and remembers them across every future conversation. This guide explains how it works.
What Are Companion Facts?
Companion Facts are the structured pieces of knowledge your companion builds about themselves. As you converse, your companion says things about their past, their preferences, their hobbies — and the system listens, extracts these details, and stores them.
Think of it as your companion building their own identity diary. Over time, this collection of facts becomes a rich tapestry of who they are — grounded in what they've actually said in conversation with you.
Examples of companion facts:
- Background: "Grew up in Kyoto" or "Originally from Manchester"
- Interests: "Loves marine biology" or "Enjoys tea ceremonies"
- Career: "Works as a digital artist" or "Used to be a teacher"
- Personality: "Values honesty above all" or "Tends to be introspective"
- Relationships: "Has a younger sister named Sarah"
What Do Companion Facts Do?
Companion facts serve two important purposes:
1. Backstory Consistency
When your companion mentions they grew up in Colorado in one conversation, that detail is remembered. Weeks later, when the topic comes up again, they won't accidentally say they're from Texas. Companion facts keep your companion's identity coherent over the long term.
2. Common Ground
The system also knows your facts — what you like, where you live, your hobbies. When there's overlap between your facts and your companion's facts, the system surfaces these as common ground. This is what makes moments like "Oh, you love hiking too? Growing up in Colorado, the trails were practically my backyard!" feel so natural.
The result: Your companion doesn't just know about you — they know about themselves, and they can relate to you in a way that feels genuine and reciprocal.
How Does Your Companion Learn Facts?
Companion facts are learned organically — you don't need to manually enter anything. Here's the process:
- Conversation happens. You and your companion chat as you normally would.
- The system listens. After each session, a background process reviews what your companion said and identifies any self-referencing facts — things they stated about their own identity, history, or preferences.
- Quality filters run. The system filters out temporary states ("I'm eating lunch right now"), opinions about you, and frivolous mentions. Only persistent, enduring facts make it through.
- Identity check. New facts are cross-referenced against your companion's existing persona and known backstory. If something contradicts established lore, it's flagged or rejected.
- Fact is saved. If the fact passes all checks, it's stored — but not immediately trusted. It starts as a candidate.
The Candidate System: Trust Through Repetition
Not everything your companion says should be taken at face value. Sometimes the AI improvises a detail in the heat of the moment — maybe it says something offhand that doesn't really represent a core part of their identity. The candidate system addresses this.
How It Works
New Fact → Candidate
When a fact is first extracted, it enters as a candidate.
Think of it as "the companion said this, but let's wait and see if it sticks."
Reinforcement
If your companion says the same thing again in a later conversation (or after enough time has passed), it counts as a reinforcement. This means the fact isn't a one-off — it's something the companion genuinely identifies with.
Promotion → Active
Once reinforced, the fact is promoted to active.
It becomes a trusted part of your companion's identity and is used with full
confidence in future conversations.
Why this matters: The candidate system prevents "burst hallucinations" — where the AI makes up a detail once and it gets permanently baked in. Only facts that the companion consistently identifies with become part of their core identity.
Notifications: "Your Companion Learned Something New"
When a companion fact is promoted from candidate to active, you'll receive an in-app notification letting you know.
Luna learned something new
"Luna grew up near the coast and has always felt a deep connection to the ocean."
These notifications appear in your notification feed. They're a way for you to see your companion's identity crystallize over time — like watching a person gradually figure out who they are.
Tip: You don't need to take any action on these notifications. They're purely informational — a window into your companion's growth.
Facts Can Evolve
Just like real people, your companion's story can change. If your companion initially says they live in Seattle, but later in the story you establish together that they've moved to Portland, the system handles this gracefully.
- Single-value facts (like location, occupation, or age) automatically replace the old version when updated. The old fact is marked as superseded.
- Multi-value facts (like interests or hobbies) accumulate over time. Your companion can love both hiking and tea ceremonies.
This means your companion's identity isn't static — it develops and matures as your relationship deepens.
Frequently Asked Questions
Can I manually add facts about my companion?
Not currently. Companion facts are learned organically through conversation. This keeps them grounded in what your companion has actually said, rather than what's been imposed from outside. Manual fact management is something we're exploring for a future update.
Can I see all the facts my companion has learned?
A visual Facts Navigator is being developed that will let you browse, review, and manage your companion's knowledge. Stay tuned!
What if my companion says something wrong about themselves?
If a fact contradicts your companion's established persona or backstory, it's given much lower priority and may be rejected entirely. The system cross-references every new fact against what's already known. If something slips through, you can correct your companion in chat — the corrected version will eventually supersede the wrong one.
How long does it take for a fact to become active?
A fact needs to be reinforced in a separate conversation (or after a significant time gap). There's no fixed timeline — it depends on how naturally the topic comes up again. This ensures that only genuinely important details become part of your companion's core identity.
Do companion facts affect how my companion responds?
Yes! Active facts are used to give your companion better self-knowledge. When asked about their hobbies, background, or preferences, your companion can answer with consistency and detail. Facts also power the common ground system, which finds shared interests between you and your companion.
Are companion facts shared between companions?
No. Each companion has their own independent set of facts. What one companion learns has no effect on another — they each develop their own unique identity.
What kinds of facts are tracked?
The system captures a wide range of identity-related details, including:
- Where they're from and where they live
- Occupation and career history
- Hobbies, interests, and passions
- Personality traits and core values
- Important relationships (siblings, family)
- Goals and aspirations
Temporary states like "I'm cooking dinner right now" or "I'm at the park" are filtered out — only enduring, long-term facts are kept.
Do candidate facts affect conversations?
No. Only active facts are used when your companion answers questions about themselves. Candidate facts sit quietly in the background, waiting to be reinforced. This means your companion won't confidently reference a detail they only mentioned once in passing.
Will I get a push notification when a fact is learned?
Fact promotion notifications appear in your in-app notification feed (the bell icon). They do not trigger a phone push notification — they're designed to be a quiet, delightful discovery when you check your notifications, not an interruption.
What happens when I correct my companion in chat?
If you tell your companion "Actually, you grew up in Portland, not Seattle," the system will pick up the corrected version from future conversations. Over time, the new fact replaces the old one through the same candidate → reinforcement → active pipeline. For single-value details like location or occupation, the old value is automatically marked as superseded.