CorvinOS
A self-hosted AI workflow OS that routes models, learns from outcomes, and records auditable decisions.
- Source repo
- CorvinLabs/CorvinOS
- Stars
- ★ 25
- Last updated
- today
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 70/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Development teams using Claude Code, Codex, or Hermes Agent that want self-hosted workflows reachable through Discord, Telegram, WhatsApp, Slack, or email.
- Not a fit if
- Teams requiring independently verified compliance certification or cost guarantees
- Users unwilling to run a local service and retain audit state
- Teams that only need a simple chatbot without a learning loop
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-20latest release tier1-complete (2026-09-18)
- Source review
- 70/100 · Some gaps
What does this agent do, and when should you use it?
CorvinOS is a self-hosted system for AI workflows that can connect Claude Code, Codex, or Hermes Agent and expose a local browser console. Its four-layer control plane consists of DataHub, Creator 2.0, the Learning Daemon, and the Dashboard, covering ingestion, skill generation, feedback learning, and operational visibility. Requests are routed among Haiku, Sonnet, and Opus, after which outcomes feed a six-dimensional loss assessment and weight-update cycle. Versioned Python skills are its execution units; they run deterministically, may call an LLM, and produce quality measurements, routing updates, and hash-chained audit events. Installers for macOS, Linux, Windows, and WSL create an isolated Python environment, add offline voice models, and start the console on 127.0.0.1. The repository presents the architecture as GDPR- and EU AI Act-compliant, although the supplied material contains no independent certification or audit.
DataHub ingests memory, files, RAG, and MCP sources, performs quality scoring and security scanning, and emits scored data manifests. Creator 2.0 generates skills through 12 structured phases while tracking loss at each phase; skills execute as Python programs, can optionally invoke an LLM, and compose through DAG-validated imports. The Smart Router assigns requests to Claude Haiku, Sonnet, or Opus, while each execution emits loss components for data quality, generation quality, user satisfaction, efficiency, learning-loop health, and system health. The Learning Daemon processes feedback, attributes outcomes to data sources, and updates weights and the routing map with a documented learning rate of 0.01. The Dashboard displays skill lifecycles, learning metrics, token allocation, and the audit chain, and produces GDPR-oriented reports. The described audit path stores metadata rather than prompts and includes consent gates, PII flagging, secret redaction, and cascading erasure with proof logging.
- Development teams using Claude Code, Codex, or Hermes Agent that want self-hosted workflows reachable through Discord, Telegram, WhatsApp, Slack, or email.
- Organizations making many Claude calls that want workload-specific allocation across Haiku, Sonnet, and Opus.
- Regulated teams that need records of model decisions, source attribution, consent, and erasure actions for audit preparation.
- Platform engineers who want repeatable AI workflows packaged as versioned, composable, DAG-validated Python skills.
- Teams needing local voice interaction in one of the documented Piper languages, with speech recognition that requires no per-language download.
- Maintainers who want a fail-closed pull, console-build, critical-E2E-test, and push cycle before updating main.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
curl -fsSL https://raw.githubusercontent.com/CorvinLabs/CorvinOS/main/install.sh | shirm https://raw.githubusercontent.com/CorvinLabs/CorvinOS/main/install.ps1 | iex
- Install notes cover
- Windows · macOS · Linux · WSL
How do you install or deploy this agent?
The quick start creates a fresh Python environment, so the source explicitly says that system Python is unnecessary. Installation requires network access to download the repository, packages, and voice models. The supplied material does not specify API keys or other credentials required for the first model invocation.
On macOS, Linux, or WSL:
curl -fsSL https://raw.githubusercontent.com/CorvinLabs/CorvinOS/main/install.sh | shOn Windows PowerShell:
irm https://raw.githubusercontent.com/CorvinLabs/CorvinOS/main/install.ps1 | iexTo install from a local checkout:
git clone https://github.com/CorvinLabs/CorvinOS.git
cd CorvinOS
./install.shFor a local checkout on Windows:
install.ps1 -Editable .\To prevent automatic Claude Code installation:
bash install.sh --no-claude-codeFor the lightweight console-only preset:
bash install.sh --preset minimalHow do you use this agent?
The installer starts the local service, verifies that the application shell returns HTTP 200, checks that local login issues a session cookie, and then opens:
http://127.0.0.1:8765/console/To permit pairing over the LAN, select that mode during installation:
bash install.sh --lanTo refresh an existing installation, rebuild, restart, and verify it:
sh update.shTo rebuild and restart without downloading new code:
sh update.sh --rebuild-onlyThe Windows update command is:
powershell -ExecutionPolicy Bypass -File update.ps1Preview uninstall targets first with:
bash uninstall.sh --dry-runA normal uninstall stops services, archives non-downloadable state such as secrets, bridge pairings, the audit chain, sessions, and voice models, and then scans for leftovers:
bash uninstall.shWhat are this agent's strengths and limitations?
- Combines DataHub, the 12-phase Creator 2.0, the Learning Daemon, and the Dashboard into an ingestion-to-feedback control plane.
- The updater verifies the new console and API and automatically rolls back failures; developer checkouts preserve and reapply local changes.
- The documented audit design covers hash-chained events, source attribution, secret redaction, PII flagging, consent gates, and proof-logged deletion.
- Installers cover macOS, Linux, Windows, and WSL, create their own Python environment, and verify the console and login before declaring success.
- Downloadable Piper voices enable offline speech output for a documented set of languages.
- The routing examples and cost model focus on Claude Haiku, Sonnet, and Opus; equivalent routing configuration for other model providers is not documented.
- The stated 60–80% savings, 2–5% quality gains per cycle, two-to-three-week convergence, and 99.9% SLA are project claims without independent benchmarks in the supplied material.
- The complete learning model requires ongoing outcomes and user feedback plus persistent weights, audit-chain data, sessions, and other local state.
- Installation and updates download code, dependencies, and models and start local services, so security-sensitive organizations should review the remote scripts before execution.
- The material does not provide concrete first-time configuration steps for model credentials, messaging bridges, or email, leaving integration effort uncertain.
How does this agent compare with similar options?
Compared with sending every request to one expensive model, CorvinOS routes work among Haiku, Sonnet, and Opus and adjusts its weights from feedback; the tradeoff is operating a learning daemon, retaining audit state, and accumulating enough samples. The source also contrasts all-Opus and all-Haiku strategies: the former is presented as high-quality but expensive, while the latter is inexpensive but lower-quality in its example. CorvinOS aims to learn a cost-quality point between them. These cost and quality figures are repository examples rather than independent evaluations.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| CorvinOS This agent | 70 · Some gaps | CLIFree + model costs | ★ 25 | today | Python | Codex · Claude Code |
| PenguinHarness | 67 · Some gaps | Desktop appFree + model costs | ★ 3.3k | today | TypeScript | OpenAI API · Claude API |
| OMA Open Multi-Agent Runtime | 79 · Good | CLIFree + model costs | ★ 7.1k | 4d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| Entroly | 75 · Good | CLIFree + model costs | ★ 488 | 3d ago | Python | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The Forge skill documents sandboxing, timeouts, read/write mount annotations, secret redaction, and auditing. The README also describes consent, data-class and egress gates, and plugin permission approval, giving concrete support for least privilege, confirmation, and data-flow transparency; several claims remain documentation rather than implementation evidence verified by this review. The security checklist claims encryption and secret masking, but labels AES-256-GCM as a phased implementation and gives limited detail, so sensitive-data handling is not fully established. The installer pins uv and verifies its checksum, but the sample does not show a complete dependency lock or vulnerability response process. Update rollback, backup, and recovery instructions are concrete, supporting full rollback credit. Marketplace metadata names an individual owner and email, but publisher identity is unverified and a clear security-update responsibility path is not shown, so attribution is partial.
The evidence has material internal conflicts: the README calls several gates LIVE, while the Phase 9 documents describe severe defects and a production block; another later note says the related managers have no production caller and the routes return 501. Dates and status statements do not fully reconcile, lowering self-consistency. Install instructions specifically describe local dependencies, checksum verification, startup checks, and rollback on failure, but the sample cannot establish ongoing availability of external services or models. The documents describe error handling, exit codes, HTTP verification, and rollback, but evidence for error-message handling is scattered and does not establish consistency across all paths.
The README explicitly covers individual and team use, multiple chat channels, console, voice, plugins, and A2A, and distinguishes live, shadow, unbuilt, and gated features; audience and scenario coverage is strong. The Forge skill defines when to use it and when not to, and plugin descriptions give sandbox and permission boundaries, but broad overview claims sit alongside documented defects and 501 routes in the sample. Skill triggers are fairly specific, although the router skill itself uses a coarse heuristic and skips multiple phases. Installation guidance covers Linux, macOS, WSL, Windows, Claude Code, ports, LAN, services, and non-interactive runs, supporting strong environment fit.
The README provides clear entry points for overview, setup, updates, uninstall, security, organizations, and extensions. Install options, platform differences, update verification, uninstall backups, and troubleshooting links are specific; examples, commands, and FAQ-like expandable sections are extensive. The README identifies unbuilt features, gated states, and the missing plugin update path, so known limitations are documented thoroughly. The root LICENSE contains the Apache 2.0 text, but some marketplace plugins are marked MIT; the product/component license boundary merits clarification, though the stated license itself is well documented. Version badges and a CHANGELOG link are present, but the sample does not include changelog contents, limiting version-history evidence. CLAUDE.md is repository guidance for Claude Code and does not clearly establish product maintenance responsibility; unverified publisher provenance does not fill that gap.
The product describes practical uses: connecting an existing CLI agent to team channels, with audit, memory, model routing, and plugin extension. Forge specifies JSON output and artifact locations, giving some usable output guidance, but the sample lacks sufficient user-facing output examples or independent outcome evidence. The README provides cost-saving figures from the maintainer's installation and a receipt script path, while stating time windows and limitations; this supports potential marginal value and cost-benefit, but the data is from the maintainer's environment and does not justify full marks.
The README links compliance, savings, and learning claims to specific documents, scripts, and implementation areas; security documents also name files, tests, and flows, making many claims traceable. However, some evidence is a self-reported security checklist, statuses conflict, and this review cannot execute verification. Multiple sources partly corroborate installation verification, rollback, gates, and auditing, while the two Phase 9 documents expose unresolved or changed status, limiting cross-source consistency. Some pages clearly label LIVE, SHADOW, NOT BUILT, and measurement windows, separating implementation state from plans; strong compliance wording and checklist pass claims could still be mistaken for independent verification, so full marks are not justified.
- The README's LIVE claims conflict with severe-defect and NO-GO/501 statements in the Phase 9 security documents. Verify current code paths before deployment; do not treat overview or checklist claims as independent security validation.
- The security checklist describes AES-256-GCM as a phased implementation. The sample also does not establish publisher identity or a complete dependency-risk response process; confirm actual secret storage, encryption, and update ownership before deployment.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
tier1-complete CorvinOS Tier 1 Complete — Production Release
Release notes excerpt
CorvinOS Tier 1 Complete — Production Release Release Date: 2026-09-18 Version: tier1-complete (v2.0.0-rc1) Status: Production Ready ✅ 🎯 Milestone: Tier 1 Completion (Phase C) This release marks the completion of Tier 1 initiatives across all 5 components: ✅ Completed Components 1. Corvin-Knowledge Claude Code Plugin MVP [ADR-0884] - Full integration with Claude Code IDE - Knowledge graph export/search - Decision traceability 2. Parallel Orchestration Framework [ADR-0688] …View on GitHub ↗ -
v1.0 CorvinOS v1.0 — Production Ready
Release notes excerpt
🎉 CorvinOS v1.0 — Production Release What's New - ✅ All 44 plugins implemented (mix of full + stub, tracked in roadmap) - ✅ 900+ tests — all passing - ✅ 3 security fixes deployed: - HIGH-1: Audit chain race condition (concurrent stress tests verified) - HIGH-2: Learning plugin tenant isolation (GDPR Art. 5 compliant) - MEDIUM-1: MD5→SHA256 cryptographic upgrade (NIST baseline) - ✅ GDPR & EU AI Act compliant (Art. 5/6/30/32 verified) - ✅ 100% backward compatible — zero breaking changes Security & Compliance …View on GitHub ↗ -
v0.10.107 v0.10.107: corvin stop — real shutdown command
Release notes excerpt
Added — corvin stop A top-level command that shuts Corvin down (console + messaging bridges), so it can be started again with corvin serve. Fixed — corvin gateway stop was silently broken It called bridge.sh stop on POSIX — not a real bridge.sh subcommand — so every call silently did nothing. On Windows it was an explicit no-op, never touching the Stufe-1 login-autostart Scheduled Task (CorvinOS-Console, ADR-0184) that's the actual persistent process there. …View on GitHub ↗
FAQ
Does the software itself cost money?
Do I need a system Python installation?
What happens if an update fails?
Does uninstall immediately destroy audit and pairing data?
Is its GDPR and EU AI Act compliance certified?
Does CorvinOS work on Windows?
irm https://raw.githubusercontent.com/CorvinLabs/CorvinOS/main/install.ps1 | iex.Is CorvinOS open source?
Is CorvinOS safe to use?
Is CorvinOS still maintained?
Who makes CorvinOS?
Thanks — we use this to improve the page.