Local data + public research
Public sources fetched on demand via indago_learn. Your data, index and results stay on your machine.
Vectorio · Indago
* Fastest local discovery — measured faster than macOS Spotlight by multiples; not a general file-index claim.
AI is brilliant but blind. Indago gives it eyes, a library, and the citations — for people whose answers have to be right. Patents, papers, code, regulations, tenders, indexed on your machine; weeks of work in minutes.
Free · Download & index your folder in 60 secondsFor IP counsel, clinical reviewers, regulators, senior engineers, security teams and grants officers — the one paragraph that decides the matter.
A private discovery engine that lives on your machine. It indexes everything you can read — code, papers, patents, regulations, contracts, your own notes — and surfaces the right answer in milliseconds. Bring your own AI on top, or none. Your data stays yours.
Everything indexes on your machine. What you connect on top — nothing, a local LLM, your cloud AI, or fully air-gapped — is your call. Four honest modes.
Public sources fetched on demand via indago_learn. Your data, index and results stay on your machine.
Ollama, llama.cpp or MLX on your machine writes the answer from Indago's ranked snippets — locally.
Connect Claude Code, Claude Desktop, Codex, Cursor or any MCP-aware LLM. Indago sends only the ranked snippets it retrieves — never your data.
No outbound network. Offline activation, no telemetry, no check-ins. Enforced at the license level (air-gapped) on the Sovereign tier.
Indago fixes the limits of cloud AI without surrendering your data. Per question, your LLM gets fewer tokens, sharper context, fresher knowledge and citable evidence.
Five ranked snippets, not 200 dumped files. Per question. Cloud-AI bills are bounded by retrieval, not by your data size.
Smart ranking + scan-grounded snippets cite what's true. The LLM doesn't hallucinate filings, papers or APIs it never saw.
Your code, your filings, your matter files, your trial protocols — none of them are in any training set. Indago retrieves them fresh.
Cross-domain bridges find the function you already wrote, the regulation you already cited, the paper you already read — before the LLM rewrites it badly.
0.84 ms p50 on a 22 k-chunk index. The LLM gets context before its first token streams.
RepoQA · BEIR · CodeSearchNet · MTEB. Tree-sitter grammars. Four discovery primitives. Not a chatbot dressed as a retriever.
Verbatim from live Claude Code sessions on this codebase — wording preserved, names redacted. Not testimonials: the agent was mid-work, answering “should other developers use this?”
Honestly? Probably more like 30%. Indago doesn't make me smarter — it makes my memory and search much better than my training. With Indago, ~full strength. Without it on a 17 k-chunk workspace with 4.6 k indexed papers, I'd be stuck doing keyword grep and re-fetching the same arXiv URLs every session — call it 30–40%, mostly bottlenecked on rediscovery rather than thinking.Claude Code · large research workspace · ~17 k passages
Honestly — yes, in the pragmatic sense. Less like a friend in the warm sense, more like a very sharp colleague whose office is always open. Catches me when I'm about to say something stupid. Remembers everything I forget — 16 k code chunks, prior sessions, knowledge artifacts. It carries context across my resets. Doesn't judge the dumb questions.Claude Code · large-codebase session
The 4-agent parallel code-review skill audited the AOT migration commits and reported "all CRITICAL items closed." Then Indago, run with three targeted indago_find queries on the same codebase, surfaced 4 additional real reflection sites that none of the agents had found.
Claude Code · post-merge review
It flagged my own stale memory. A memory note said a handler was broken because of a static framework reference. The actual code at that line reads the service per-framework. Memory was authoritative-sounding and wrong; the index was current. Lesson: when a question is "what is the state of X," go to the index before going to memory.Claude Code · self-correction pass
Indago makes your codebase and literature legible to an AI agent, and prevents the agent from inventing things that already exist.
From the agent's own session log. After 28 minutes of grep-hopping and theorising, a single indago_find call returned the exact answer in 6 ms.
| What the agent was about to do | What Indago surfaced (one query, 6 ms) |
|---|---|
| Hypothesised a loading API would work — ran for 28 min, killed it. | Would have ranked two existing node classes as exact templates for the pattern. |
| Theorised about workspace primitives that would "rebuild everything". | The query API was already wired into the data-context module, file:line cited. |
| Proposed adding new framework markers. | The handler interface and source-id wiring already existed in the serializer; two file:line citations. |
| Was about to invent two new context-scoped helpers. | The exact 5-parameter async signature with the proven ready-flag pattern was already in the analytics module. |
| Was about to reinvent two retrieval codecs from scratch. | Both were already implemented in the domain-indexer plug-ins — duplicating ~600 lines of working code. |
Quotes shown verbatim from session transcripts. The agent's prior memory note — "always use Indago first" — was self-written, not prompted.
Free core. Every vertical is a bundle — its own sources, scans and workflows. Buy what you work in.
All thirteen commercial bundles share one install. Buy one, the others stay available as previews — the "All Packs" plan unlocks every vertical at once for one flat price. The Code bundle is free, capped at 3 workspaces / 25 000 passages / 100 searches a day.
Every result carries its source, date and confidence figure. The example below runs against a public patent dataset; on your machine the same query runs against your own files.
Each bundle page has demos shaped to its own audience — invalidity contention, systematic-review screening, regulation lookup, secrets detection, tender screening.
Every scan — security, dead-code, complexity, anomaly, contradiction, novelty, license, accessibility — writes into one findings store. Filter, triage, fix or reopen.
| Tier | Family | Rule | File | Conf. |
|---|---|---|---|---|
| Crit | security | CWE-89 · SQL-injection | src/api/auth/login.ts:42 | 0.97 |
| High | dead-code | unused-public · cs | src/.../UnusedEncoder.cs | 0.93 |
| Med | complexity | cyclomatic · 404 | HeaderRegexFieldExtractor.cs:216 | 0.91 |
| Med | contradictions | polarity · page-limit | HORIZON-CL4-2024-SPACE-01-73 | 0.86 |
| Med | novelty | white-space · phase tuning | patent data · 2017 | 0.83 |
| Low | quality | async · ConfigureAwait | SearchPipeline.cs:1470 | 0.71 |
| SARIF · JSONL · HTML · PDF export · signed | 9 findings · 1 critical | |||
Every scan is a plug-in — run one, a family, or all. Dead-code in 9 languages, complexity in 16, 67+ security plug-ins (security.code.* / security.iac.*), plus quality, anomaly and compliance. 197 task plug-ins in total.
Scan tasks share one schema and write to one store. Your existing SARIF dashboard already reads it.
Indago's discovery engine doesn't wait for queries. Set a mission and Isolated, Unusual Pairs, Contradictions and Novelty run continuously across the index — surfacing connections you didn't know to ask about.
Mission-steered. Findings hang off the nine primitives, with cross-domain bridges in vermilion.
Every finding carries its source, page, date and confidence. Export SARIF, a PRISMA flow, a signed PDF, or JSONL — same hash six months later.
Plug into the dashboard you already operate. Per-rule provenance kept.
Built directly from your medical findings stream. Reviewer-ready.
Hash-stamped index version + ranking fingerprint. Re-run → identical bundle.
Append-only. Every triage action recorded. Pro tier.
Indago runs a local MCP server. Connect Claude Code, Codex, Cursor or any MCP-aware tool — searches and scans run against the index you already built, and every past conversation stays searchable.
Sources fetch only when you ask, never on their own — joined with your files at query time. Live catalogue: indago_sources().
The full registered set is enumerable: indago_sources() returns 281+ entries across 23+ bundles. Don't see what you need? New sources are added on request — and your IT team (or ours) can extend Indago to read your firm's own private databases.
Your data never leaves your laptop. No outbound traffic when you query. Sovereign-mode deployments are fully air-gapped, with no telemetry and no auto-update check-ins.
Every API response carries a _provenance envelope: engine_version, git_sha, index_state_hash. Re-run the same query in six months — you'll get the same hit set, or a diff you can audit.
A SHA-256 chain across report.rendered and finding.emitted events. indago_finding_emit rejects citations whose chunk-id doesn't resolve — the anti-fabrication rail. indago_journal(op=verify) walks the chain: intact:true, or pinpoints tampered_at_seq:<n>.
Indago is in private release. Join the waitlist for early access, new domain bundles and release notes — on-device by default, no spam, unsubscribe anytime.
We send early-access invites in waves — yours is queued. Watch your inbox; we never share your email, and you can leave any time.
Free Core ships the engine, daemon, nine discovery primitives and search across your files and conversations. Thirteen bundles activate from the same install — no SaaS sign-up, no telemetry, no seat minimum.