Nimbus

Nimbus is an open-source, local-first AI agent designed for engineers operating systems in production environments, positioned as the on-call intelligence layer for DevOps, SecDevOps, and platform engineering teams. It runs a headless Nimbus Gateway on your machine, maintains a private SQLite index across the entire developer toolchain (source control, CI/CD, cloud infrastructure, monitoring, incident management), and executes multi-step tasks on your behalf.

Key capabilities

• Local indexing and fast retrieval: searching 50,000 indexed items across 5 services takes less than 100 milliseconds, and most queries require no network access.
• Cross-service context correlation: answer who wrote a given line of code, when, and in which commit, and can extend to correlate the PR that carried the change, the ticket that requested it, and the incident it triggered.
• Broad connectors: provides 90+ first-party MCP connectors, covering Google, Microsoft, GitHub/GitLab/Bitbucket, Slack, Jira, Linear, Notion, Confluence, as well as observability, CI/CD, security and quality, feature flags, GitOps, data and BI, deployment, and finance and support tools.
• Bidirectional MCP support: acts as an MCP client to drive connectors; as an MCP server, exposes the local index and built-in agent through 21 read-only tools (9 indexing tools + 12 agent tools, such as explainWhy, findExpert, assessImpact, getCatchup) to any MCP client; when [mcp_sink] is enabled, it can additionally provide an optional write tool, capture_note.
• Security and consent mechanisms: credentials are stored in the operating system's native keystore (Windows DPAPI, macOS Keychain, Linux Secret Service) and are not written to disk, logs, or IPC responses. All destructive or outbound operations are intercepted by HITL gating at the executor layer rather than relying on prompts, and cannot be bypassed by jailbreaking. Connectors run in sandboxed subprocesses and receive only the credentials for the services they declare. Outbound operations are recorded in an append-only egress ledger based on a BLAKE3 chain, which can be reported with nimbus prove. Prompt injection is mitigated by injecting file contents and API responses as typed <tool_output> data blocks.
• Cross-platform: Equal support for Windows, macOS, and Linux; every PR runs the full gate on Ubuntu, and pushes to main run the three-platform matrix in parallel. Supports OS desktop notifications.
• Demo mode: nimbus demo can seed synthetic "Acme" organization data (people, PRs, incidents, deployments) into an isolated demo root directory without touching real configuration or data; nimbus demo stop stops the demo gateway, and nimbus demo reset deletes the entire demo root directory.

How to use

1. Install: Supports native installers, GPG-signed apt/yum repositories, AppImage, portable archives, and package managers (automatic updates). The install script verifies release signatures. macOS and Windows require no administrator privileges; Linux .deb uses sudo. Without sudo, .deb or AppImage is available.
2. Index repositories: Run nimbus init to add repositories to nimbus.toml and enable code indexing, start the gateway, and complete indexing. No account, token, or API key is required.
3. Trace code provenance: nimbus why can answer author and time based on local git history; after connecting GitHub, issue trackers (Jira or Linear), and incident tools, the same command can be extended to PRs, issues, and incidents.
4. Optional LLM integration: Indexing, nimbus why, and agent briefings all work without an LLM; an LLM can provide natural language queries (nimbus ask) and narrative synthesis ([agents] synthesis). It can point to a local model for fully offline operation.
5. Optional cloud service connections: Use nimbus connector add or nimbus connector auth <service> to add connectors.

Use cases Engineers and operations staff whose on-call rotations span multiple monitoring tools and cloud consoles; DevOps, SecDevOps, and platform engineering teams that need to quickly locate incident context, code provenance, PRs, deployments, and DORA metrics across services; and teams that need local-first operation with credentials never leaving their environment.

Limitations

• There is no managed cloud service, Nimbus account, or relay server.
• Extensions are currently not executable: nimbus extension install copies, hashes, verifies the Ed25519 signature, and records it, but does not start installed extensions, so installing extensions does not give the agent new capabilities; what actually runs are about 90 first-party connectors.
• The local extension marketplace is code-complete in the Tauri desktop app but has not yet been released, and is planned for Stage 13 as the desktop v0.1.0 tag.
• On headless Linux (servers, containers, SSH sessions, or WSL), libsecret requires a D-Bus session and an unlocked keyring, which are usually missing; run nimbus doctor to see what is missing.
• On Windows, network authorization for third-party MCP servers is all-or-nothing.
• The project uses stages rather than calendar dates; Stages 1-6 are complete; starting from Stage 7, it is built according to the Sequencing Spine overlay (S1->S5).
• Licensed under AGPL-3.0.

Pricing and charges: No relevant information is currently provided in the materials.