How scattered data becomes useful answers.

A canonical shape (kind, account, record id, title, excerpt, observed-at, and deletion behavior) stored on the row itself, not bolted on later. The exact-source pair (account + message) is a runtime-enforced invariant.

Unverified inbound claims land in a holding table. Promotion to durable memory requires user-authored, trusted-contact, or corroborated status, and rejected claims are suppressed so they aren't relearned.

Delete a message and the data derived from it tombstones or cascades. Writers of source-backed data register deletion observers so stale citations can be cleaned up instead of quietly hanging around.

Embeddings handle "the invoice from Acme last quarter." BM25 handles "ORD-4821." Results are fused so phrasing and exact match both work. Browsing and name lookup keep working even with no LLM or embeddings available.

An assistant question blends timeline events, context facts, the knowledge-graph neighborhood, and direct message evidence, so receipts, charge amounts, deadlines, and dispute history come from real records, not recall.

Ambient detection can notice meeting-like activity, but recording is explicit. The capture lane records metadata about the session and stores transcript segments as source-backed material.

The default path uses a local Whisper model from the model cache, with configured cloud providers available as fallback routes when the user enables them.

Human-readable export turns communication memory into Markdown notes with source references, so users can audit, archive, or move selected context without reverse-engineering a database.

Tools carry a declared contract (parameters, return shape, a sample input) and compile through Roslyn inside a shared sandbox. A safety gate blocks raw HTTP and filesystem access; all web egress goes through one auditable channel. Compiles are cached; execution is time-bounded.

Each tool stamps its last run: when, ok-or-error, and the output. Nothing is an opaque blob: an "Inside" workbench shows the code, the contract, and the last result, and lets you edit or run any tool by hand.

Once an agent has authored a bulk-fetch tool, repeat refreshes run the compiled tool directly: a 21-symbol portfolio refresh dropped from ~42 model calls to zero. If the tool breaks, the system self-heals back to the model to repair it.

Hiring is stage → approve → activate: the assistant provisions the worker's profile and connection-tested email account, then waits on a human approval task. Credentials never ride approval payloads; pasted secrets are redacted from transcripts before validation. Workers are machine-bound and never sync.

Peers advertise the local models they can serve; a remote-peer LLM provider forwards a completion to the machine with the hardware. A load-scored allocator hands scheduled tasks and background jobs to the least-busy node.

Journaled replication with per-origin high-water marks and LWW conflicts; snapshot bootstrap for new peers; automatic gap healing after long offline stretches; large files stream in resumable, hash-verified chunks. Short-TTL work claims stop two machines paying to process the same message.

Add an MCP server such as filesystem, GitHub, Slack, SQLite, browser search, or an internal tool. Its tools land next to the built-ins, and the assistant picks them through the same routing and approval flow.

Point your own agent (Claude Code, Codex, or another MCP client) at Talavine's built-in server and query the same source-grounded data, with citations, behind an explicit external-use approval gate.

Scans for a local Ollama install and registers detected models. Local-first when local works, cloud when it doesn't, and a fully local option for private work.

Classification on a fast model, summaries on a smart one, high-stakes drafting on a frontier model. Each function type picks its own tier, and LocalOnly is enforced before any cloud advisor runs.

Hard caps per model, per agent, per day. Spend Guard cuts off the bill before it surprises you and alerts you when it does.

Prompt and model changes run through the message-pipeline calibration harness. Results are recorded by prompt hash and hardware key: the same model on different hardware is a different decision, and routing fails closed without passing proof.

A local maintenance pass marks stale claims, escalates contradictions, prunes low-confidence projections, and journals an auditable insight row. Conflicts surface for review instead of being silently merged.

When enrichment marks a message not knowledge-worthy, durable artifacts are dropped before persistence. Ungrounded quotes are dropped. The model classifies; the system gates the durable writes.

A background steward that studies every model you could use and refuses to route without function-scoped, hardware-keyed benchmark proof, plus a serving-health lane that fixes the local runtime, not just the model choice.

One calibrated model call per message, often zero, with production replay harnesses, champion/challenger evaluation, junk economics, and a phishing detector that patches itself.

Provenance on every fact, one ingestion funnel for every source, a local knowledge graph with pre-built neighborhoods, and trust gates between "a message said it" and "the system believes it."

The sandboxed C# substrate under Vibes and the generated panels: Roslyn compilation that works even on mobile, a semantic safety analyzer, one audited egress channel, and tools that outlive the model calls that wrote them.

Journaled peer-to-peer sync with self-healing catch-up, ECDH-authenticated LAN crypto, a load-scored compute pool, and supervised worker processes, with no public gateway anywhere in the design.

Every consequential action runs one mandatory lifecycle and leaves a durable receipt, so when you ask "why did you do that?", the assistant answers from evidence: who acted, what triggered it, what changed, and how to undo it.

The full architecture story in one document, written for technical agent builders: data plane, assistant runtime, Model Scout, context engine, MCP, and the trust model.