Three layers, plainly. Augment your agent with no key, orchestrate a governed SDLC on your own keys, and keep both on a shared platform that survives every session.
Helixor Code runs as an MCP server. When your agent needs to find or understand code, it queries the structural Code Index — not grep — and reads and writes stateful TODOs. No key is required because your agent supplies the model.
For nontrivial or regulated work, Helixor Code drives a workflow from the catalog and routes each step to the right model on your key — Claude triages, a human approves, Codex implements, with audit evidence at every gate.
Indexes and TODO stacks live server-side, so they persist across sessions and machines and can be shared with permissions. Built on the Helixor engine — works with any LLM, or without one.
When an agent starts with grep or file search, it has to read dozens of candidate files into its context window, exhausting token limits and ballooning latency. Helixor parses source trees across 12 languages using Tree-sitter into a persistent graph of symbols, call chains, data flow, and architecture rules.
The agent calls code.search and gets exact definitions, usages, and weighted bonds in a compact ~150 token payload rather than thousands of tokens of file dumps.
When files are modified in git, Helixor splices working diffs into the index. The agent knows before answering if a file has changed.
Ad-hoc coding agent prompts can drift, hallucinate edge cases, or introduce subtle architectural flaws. Helixor Enterprise orchestrates change delivery through checkpointed, deterministic workflow graphs where each stage produces an auditable artifact.
Enterprise security requires that proprietary source code never leaves the developer's authorized workstation. Helixor Code decouples the code index from team memory using a hub-and-spoke architecture.
Repositories are parsed and indexed locally using native binaries. The structural database resides on the developer's disk (e.g. .helixor/), never transmitted over the wire.
Durable TODO stacks, workspace channels, architectural decisions, and server-derived identity are synced through a lightweight hub that deploys in 5 minutes and scales to zero.
Clients cannot forge author identity. Inbound ports and VPNs are not required. Complete airgap and LAN-only operation are fully supported for defense and regulated environments.
Follow the 4-step onboarding guide. Takes about five minutes with zero keys required.