Visca Agent — a personal autonomous agent.
Memory you own. Credentials you trust. Audit you can read.
One of the eight Visca products. Runs on the open Lattice Runtime.
| Area | Summary |
|---|---|
| Memory | Holds what matters. Forgets what doesn't. Structured memory — identity, preferences, projects, events — extracted with source attribution and deduplication. Hybrid retrieval (dense + sparse) ranks results semantically and lexically, with staleness windows per memory type. Per-user and per-channel isolation. Embeddings run locally by default. |
| Identity | Becomes its own. Behavior lives in SOUL.md; during onboarding the agent observes how you communicate and writes its own persona files. A per-user journal captures its reflections on past interactions. NOW.md is the ephemeral scratchpad for current focus. |
| Proactivity | Reaches out when something matters, without being asked. Every hour it re-reads its notes, notices what's unfinished or due soon, and sends a message if needed. Notifications route to the right channel and yield if you're already in conversation. |
| Trust | Fail-closed by construction. Actor identity is resolved once (guardian, trusted, or unknown) and enforced everywhere. Untrusted actors cannot read memory, write memory, trigger tools, or escalate. Credentials live in an isolated process and never reach the model. Every tool runs in a sandbox. |
1. Install
# from source (recommended during the rebrand)
git clone <repo-url>
cd visca-agent
./setup.sh2. Hatch
vellum hatchThe CLI binary is still named
vellumfor the moment — renaming it is a tracked follow-up so install scripts and shortcuts keep working through the rebrand. The product is Visca Agent.
3. Pick a mode
- Local — everything runs on your machine.
- Cast-backed (rolling out) — credentials, audit, and config sync to a Visca Cast workspace. Replaces the legacy hosted control plane.
Common commands
vellum wake # start services
vellum sleep # stop services, keep data
vellum client # interact through the terminal
vellum ps # view running assistants
vellum terminal # open a shell into a managed agent container
vellum upgrade # upgrade to latest versionAll commands target the default agent. If you have multiple, pass the agent ID as the second argument.
| Visca surface | What it gives this agent |
|---|---|
| Cast — fleet as code | The agent's canonical record, credential handles, config blob, and event/audit stream live as typed records in a Cast workspace. See cast-modules/visca-agent/. |
| Sigil — cryptographic identity | The agent presents a Sigil identity to other services. Long-term: bundle + harness + model are bound under one identity. |
| Warrant — zero-trust access | Tool calls run under short-lived, capability-scoped credentials. Replaces long-lived API keys in env vars. |
| Chronicle — knowledge graph + audit | Every memory write, tool call, and credential materialization hash-chains into Chronicle. Operators can answer what did the agent do, why, and on whose authority. |
The runtime is unchanged where it shouldn't change. What changes is the control plane behind it: telemetry, credential materialization, and config no longer call a hosted cloud — they call a Cast workspace your organization owns.
See ../visca-docs/BIBLE.md for the canonical product reference and ../visca-docs/WHITEPAPER.md for the executive distillation.
Telemetry is off by default. To enable it, set both:
VISCA_TELEMETRY_ENABLED=1
SENTRY_DSN_ASSISTANT=<your-org-dsn>Either one missing → no events ship. See assistant/src/config/env.ts for the kill switch.
We welcome contributions — bug fixes, new features, documentation improvements.
- Development: See
CONTRIBUTING.md. - Code of Conduct: See
CODE_OF_CONDUCT.md. - Security: See
SECURITY.md.
MIT — see LICENSE. Integration logos from Simple Icons, licensed CC0 1.0.
This repository is a Visca-branded fork of the open-source Vellum Assistant project, MIT-licensed; provenance is preserved in the git history.
Visca · The Autonomy Cloud · visca.ai