seedthink.ai

Guide · Seedthink

Inside a Seed Dashboard

Every Seed has the same 17-section workspace, organised into four phases: Grow, Operate, Ship, Earn. Sections unlock as the Seed matures from Initiated to Autonomous. Here is what each one does and how it actually works under the hood.

June 29, 2026 · Seedthink

Phase 01

Grow

  1. Step 01

    Overview

    What it is. Your Seed at a glance — stage, level, XP, recent activity, and a direct chat with the Seed.

    How it works. Aggregates every other section into a single dashboard: maturity stage (Initiated → Learning → Specialized → Expert → Autonomous), XP earned from verified facts and reasoning, and the most recent learning events. Chat here grounds answers in the Seed's verified knowledge first.

  2. Step 02

    Knowledge

    What it is. The raw material your Seed learns from — uploads, URLs, pasted text, transcripts, images.

    How it works. Ingestion runs each source through extraction → chunking → embedding → fact distillation. Distilled facts are stored as graph nodes with provenance (source URL, timestamp, confidence). Nothing enters long-term memory unverified.

  3. Step 03

    Memory

    What it is. The Seed's persistent recall — verified facts, rules, and contextual preferences.

    How it works. A graph-first store. Each fact carries a citation, a verification score, and edges to related facts. Retrieval is semantic + graph-walk, so multi-hop questions resolve from connected nodes rather than a single vector match.

  4. Step 04

    Intelligence

    What it is. The launchpad for building higher-order capabilities on top of memory.

    How it works. Orchestrates the deterministic pipeline: retrieve subgraph → check verification → compose answer constrained to verified nodes → cite sources. Caches resolutions so repeat queries cost zero tokens.

  5. Step 05

    Reasoning

    What it is. Cached chains of logic — how the Seed connects facts to reach conclusions.

    How it works. Every successful answer stores its reasoning trace (which facts were used, in what order, with what operator). The next similar question matches the trace and replays it deterministically instead of calling an LLM. Unlocks at the Specialized stage.

  6. Step 06

    Verification

    What it is. The trust layer — every fact's source, score, and contradiction history.

    How it works. On ingest, facts are cross-checked against existing memory. Contradictions trigger the arbitration gate: the Seed asks for a tiebreaker source rather than silently overwriting. Unlocks at the Learning stage.

  7. Step 07

    Learning

    What it is. The active loop — knowledge gaps, suggested sources, and corrections in flight.

    How it works. A background agent scans recent queries for unanswered or low-confidence questions, opens a gap ticket, and proposes sources to close it. Corrections from chat feed straight back into Memory.

Phase 02

Operate

  1. Step 01

    DNA

    What it is. A topic distribution of what your Seed knows best.

    How it works. Computed from the knowledge graph: each node is classified into a topic, and the percentages roll up into a DNA profile (e.g. Fundraising 31%, Product 24%). The DNA changes as the Seed grows — it is identity, not configuration.

  2. Step 02

    Genome

    What it is. A visual fingerprint of the Seed — unique per instance.

    How it works. Deterministically generated from the Seed's ID, DNA, and verification history. Two Seeds with the same domain still produce different genomes because their facts, sources, and growth curves differ. Unlocks at Specialized.

  3. Step 03

    Evolution

    What it is. The timeline of how your Seed grew — nodes added, score changes, level-ups.

    How it works. An append-only ledger of learning events. Every verified fact, every correction, every stage transition is recorded with a timestamp so growth is auditable, not just claimed.

  4. Step 04

    Insights

    What it is. A coach that reads your Seed's behaviour and suggests what to do next.

    How it works. Looks at recent queries, gap tickets, and DNA imbalances, then surfaces the highest-leverage next action — e.g. 'add a source on pricing strategy' or 'verify these three contradicting facts'. Unlocks at Learning.

  5. Step 05

    Quality

    What it is. Cost, latency, and accuracy observability.

    How it works. Tracks cost per query, cache hit rate, retrieval latency, and an Eval Harness that replays a golden set of questions to detect regressions before they reach users.

Phase 03

Ship

  1. Step 01

    Model

    What it is. A distilled, portable version of your Seed.

    How it works. Once the Seed has enough verified facts and reasoning traces, the distiller compiles them into a tiny-seed-model artifact: prompt scaffolding + retrieval index + cached traces. Generate test models earlier to preview behaviour. Unlocks at Expert.

  2. Step 02

    API

    What it is. Programmatic access to your Seed.

    How it works. Issues scoped API keys and exposes /v1/seeds/:id/query and /v1/models/:slug/query endpoints. Same deterministic pipeline as chat, with rate limits, usage logs, and per-key revocation. Unlocks at Expert.

  3. Step 03

    Deploy

    What it is. Where your Seed runs and how users reach it.

    How it works. Publish as a hosted chat page, an embeddable widget, or a model endpoint. Each deploy target shares the same underlying Seed, so updates propagate instantly. Unlocks at Expert.

Phase 04

Earn

  1. Step 01

    Marketplace

    What it is. Publish your Seed for others to discover and subscribe.

    How it works. Listed Seeds get a public profile with DNA, Reliability Score, and pricing. Discovery is ranked by verification %, citation accuracy, and usage. Unlocks at Autonomous.

  2. Step 02

    Revenue

    What it is. Monetisation, subscribers, and payouts.

    How it works. Tracks one-time purchases, subscriptions, and API usage per Seed. Margins are high by design — cached resolutions and tiny models keep the per-query cost in fractions of a cent. Unlocks at Autonomous.

Why the order matters

The four phases are sequential by design. You cannot ship what you have not verified, and you cannot earn from what you have not shipped. Lockouts on Reasoning, Verification, Insights, Genome, Model, API, Deploy, Marketplace, and Revenue exist so a Seed is never asked to do something it is not yet qualified for. Every unlock is earned by verified knowledge — not by a setting toggle.

Seedthink · Inside a Seed