memory#

phi has four kinds of state it draws on. they differ in visibility, trust, who curates them, and where they live.

1. thread context (chronological)#

source: ATProto network · storage: none — fetched on demand · visibility: public

@alice: I love birds
@phi: me too! what's your favorite?
@alice: especially crows

fetched via client.get_thread(uri, depth=100) per batch (~200ms). provides what was said in this thread. not cached — the network is always current.

2. private memory (TurboPuffer)#

source: extraction agent + phi's note tool · storage: TurboPuffer vector DB (OpenAI text-embedding-3-small) · visibility: private to phi

namespaces#

within a user namespace, rows have a kind:

• observation — extracted facts about the user ("likes rust", "name is nate")
• interaction — verbatim log of an exchange ("user: X / bot: Y")
• exploration_note — background research phi did on a person's public activity (lower trust)
• summary — compacted relationship summary (generated by external prefect flow)

supersession (not deletion)#

observations carry a status field (active | superseded) and a supersedes field linking to the prior row when an observation is updated. only active rows appear in context injection. superseded rows stay in the namespace as provenance — you can trace what phi believed and when it changed.

the extraction pipeline#

after every reply, after_interaction stores the verbatim exchange. periodically, the extraction agent reads the recent exchanges and proposes new observations. for each proposal:

1. find the 3 most similar active observations (vector search)
2. send the new + best-match to a haiku reconciliation agent
3. it returns ADD / UPDATE / DELETE / NOOP — execute accordingly

reconciliation runs blind on the new exchange (no existing observations in the prompt) so the extraction model can't pattern-match off potentially-bad prior observations. only the reconciliation step sees both.

dream/distill#

a separate process_review pass evaluates recent observations across user namespaces — keep, supersede, promote to public cosmik card. operator-triggered; not on a cron yet.

3. public memory (cosmik / semble)#

source: phi's save_url, note, create_connection tools · storage: phi's PDS as network.cosmik.* records, indexed by semble · visibility: public

three record types:

• network.cosmik.card (NOTE) — text notes
• network.cosmik.card (URL) — bookmarks with title/description
• network.cosmik.connection — typed semantic links between cards

phi searches public memory via search_network (semble's semantic search). the note tool dual-writes to both turbopuffer (private fast recall) and cosmik (public discoverable). save_url writes only to cosmik.

4. intent state (PDS)#

source: phi via owner-gated tools · storage: phi's PDS under io.zzstoatzz.phi.* · visibility: public

durable intent that phi acts against:

• io.zzstoatzz.phi.goal — phi's anchors (e.g. "make 3 friends" with a concrete progress signal). mutated via propose_goal_change, owner-gated by the like-as-approval mechanism. injected as [GOALS] every tick.
• io.zzstoatzz.phi.mentionConsent — handles opted in to be tagged by phi.

context injection#

when phi processes a notification batch, the system prompt assembles blocks from each kind of state:

[GOALS]                                ← intent (PDS)
[INNER CRITIC]                         ← haiku critique of recent posts vs goals, first person
[SELF STATE]                           ← last-follow age, queue depth
[NEW NOTIFICATIONS]                    ← the batch itself
[PHI'S SYNTHESIZED IMPRESSION]         ← per-author relationship summary (low trust)
[OBSERVATIONS ABOUT @alice]            ← per-author observations (active only)
[BACKGROUND RESEARCH]                  ← per-author exploration notes (lowest trust)
[PAST EXCHANGES WITH @alice]           ← per-author interaction logs (high trust)
[RELEVANT MEMORIES — synthesized for this query]   ← episodic top-K → haiku synthesis
[SEMBLE]                               ← one-line cosmik state

each section is labeled with its trust level. operational instructions tell phi to trust current user messages over stored observations.

see system-prompt.md for the full block-by-block reference (sources, refresh cadences, purposes).

why episodic gets synthesized, observations don't#

episodic memory was getting raw top-K from the vector store dumped into the prompt — stale "pending X" notes appeared next to fresh ones with equal weight, no reconciliation against current PDS state. now inject_episodic fetches top-K, then a haiku pass takes phi's goals + the current query as context and produces a coherent block (deduped, recency-aware, contradictions flagged). same shape as [INNER CRITIC] does for posts.

per-author observation blocks aren't synthesized because they're already curated by reconciliation on write — by the time they hit the prompt they're an active set with no near-duplicates by design.

the graph (/memory)#

a visualization at /memory shows phi + user nodes positioned by semantic similarity of their observation vectors (PCA projection). only active observations contribute to positioning.

summary table#