OpenPlanter: Recursive AI Investigation Agent for Entity Resolution and Connection Mapping

You've spent hours cross-referencing FEC records, state corporate registries, and federal contracts databases. You know the pain. Entity resolution is a manual nightmare. Connecting "Smith Holdings LLC" in a USAJOBS contract to "Smith Holdings" in an ORCA lobbying disclosure to "S. Holdings" in an OpenSecrets donor record takes days of work.

Analytical software can help, but it's pricey and requires enterprise licensing.

OpenPlanter claims to be a Palantir alternative for community use. What it does is more focused and realistic. OpenPlanter is worth checking out if your use case fits.

What OpenPlanter Is

OpenPlanter is an open source recursive AI agent. It runs in a terminal. The target users are analysts who need entity resolution and dataset connection mapping on a budget, no Palantir or Nuix.

It ingests structured datasets: corporate registry exports, campaign finance CSVs, lobbying disclosures, government contracts. The AI resolves entity identities across these datasets. Then it surfaces connections that manual cross-referencing would miss.

The model is autonomous. You ask a question, point to datasets. The agent handles the investigation. It uses 19 tools to organize the workflow: ingest, shell, file I/O, web search. The agent decomposes complex tasks into smaller ones.

Autonomous doesn't mean accurate. It means the agent decides what to investigate next based on findings. That's both valuable and risky.

The 19-Tool Architecture

The tool set covers the full investigation workflow in a single agent. Dataset ingestion tools handle CSV, JSON, and similar formats from public record sources. They normalize entity identifiers across datasets with different naming conventions, ID schemes, and field structures. This step usually requires manual cross-referencing effort.

Shell execution and file I/O tools give the agent access to the local filesystem and command line. The agent can run external scripts, query local databases, process command outputs, and integrate with other tools on the same machine. This capability operates with significant system access. You need to review this access carefully before using the agent with sensitive material.

Web search integration extends investigations beyond local datasets. When a connection appears in offline records, the agent can query the web to fill gaps. Real-time public records, news, and corporate filings become part of the investigation chain.

Planning and delegation tools handle investigation decomposition. Complex questions get broken into focused sub-tasks and delegated to sub-agents. Each sub-agent has its own scope and tool access. You manage the workflow this way.

The tool set includes dataset ingestion tools, shell execution and file I/O tools, web search integration, and planning and delegation tools. The agent handles CSV, JSON, and similar formats from public record sources, normalizes entity identifiers across datasets. The agent runs external scripts, queries local databases, processes command outputs. The agent queries the web to fill gaps, accesses real-time public records, news, and corporate filings. The agent breaks complex questions into focused sub-tasks, delegates to sub-agents. The agent manages the workflow.

Introduction No frontmatter was provided to preserve.

Entity Resolution Across Heterogeneous Data

Entity Resolution: The Investigative Bottleneck

Investigative teams hit a wall with multi-source public records at scale. Records come from different agencies, each with its own data standards and quality levels. A single entity can appear as multiple variations across datasets.

The Problem

Investigative teams face a significant challenge with manual reconciliation across thousands of records. A holding company might be listed as ABC Holdings Inc., ABC Holding Company, ABC Holdings LLC, ABC Inc. A lobbyist's employer might be listed under parent company name or subsidiary name. This process is a significant time-sink.

OpenPlanter's Solution

OpenPlanter's entity resolution layer maps these inconsistencies to unified entity representations. The output is an entity graph where connections between records are established through inferred matches.

The Output

The output provides connections backed by evidence, with citations to specific source records. Analysts can trace relationships back to underlying records, verify resolution logic, and assess connection validity.

Black-box conclusions aren't usable in journalism or formal research. OpenPlanter's design provides citeable, traceable connections.

Recursive Sub-Agent Delegation

The recursive delegation architecture sets OpenPlanter apart from standard data processing scripts. Complex investigation questions get broken down into sub-tasks. The coordinating agent identifies entities tied to the target, maps their political contributions, cross-references contracts, and pulls relevant news.

Each sub-task goes to a sub-agent with its own scope and toolset, consisting of 19 tools. Sub-agents work alone, querying, ingesting, searching, and sending findings back to the coordinating agent.

The coordinating agent pieces together the findings and decides what they mean for the next step. The analysis either ends or new sub-tasks are delegated based on the results.

This recursive structure allows investigation depth that’s hard to manage manually. A connection found in step three can trigger a new chain of delegations. The agent follows where the evidence leads. You still need to judge if the connections are real or just errors.

Practical Investigation Workflows

Corporate influence mapping shows what this tool can do. Ingest a campaign finance dataset. Ingest a government contracts dataset for a sector or region. Pick a company or individual to start with. The agent maps donation patterns across finance records. It identifies contract awards to the same entity and connected organizations. It cross-references lobbying disclosures for the same actors. It produces a structured analysis of relationships with citations.

Completing this work manually takes a team of journalists days, and they can only handle a single company.

Government contract analysis benefits from entity resolution. Contract awards often go to subsidiaries, joint ventures, and related entities, which share beneficial ownership with a parent organization. These are recorded under different names; the agent resolves naming variations. It surfaces related-party awards that manual review misses.

The workflow for journalists is simple. Define the investigation question, provide dataset paths, and specify the target entity. Review the structured analysis. The output is a research foundation that includes source citations, such as lobbying disclosures, campaign finance records, and government contracts. It is not a finished article but the connection map and evidence base on which a reported story is built. Journalists save time.

Limitations and Operational Considerations

Dataset quality is everything. Entity resolution inference relies on source data. Bad data means bad output. Records with inconsistent formatting, incomplete information, and missing key fields produce gaps and false links. The agent resolves entities confidently across bad data, making connections that don't verify. Every significant connection needs verification against primary records.

The agent can execute shell commands and read/write files on the local system, which is meaningful access. Be cautious deploying OpenPlanter with sensitive data, such as confidential documents, unpublished files. Review execution logs to know what commands are run and what files are accessed. You shouldn't trust the agent with sensitive data without understanding its actions.

The terminal UI and self-hosted model require technical skills. Analysts used to GUI tools like Maltego or Palantir will face a learning curve. This is not a criticism, but a description of the target user. The tool is built for those comfortable with technical setups.

Verdict

OpenPlanter fills a gap. Cross-dataset entity resolution at investigation scale has no free alternative. Commercial tools exist, but are priced for government and large newsrooms. OpenPlanter targets independent researchers, small teams, and journalists.

The Palantir comparison is natural. Palantir has decades of engineering. Entity resolution is validated at scale. Data infrastructure is managed. OpenPlanter can't replicate that. OpenPlanter is open source and self-hosted.

OpenPlanter delivers autonomous investigation depth. It provides evidence-backed connection mapping. The recursive sub-agent architecture follows chains without manual direction.

Investigations with large datasets and complex connections are a good fit for OpenPlanter. Manual cross-referencing is not sufficient. Commercial platforms are too expensive. Deploy and evaluate OpenPlanter.

Autonomous depth is key. OpenPlanter offers more than just aggregation - that's standard. Agents pursue leads based on findings. No step-by-step guidance is needed.

OpenPlanter is best suited for journalists, data journalists, and researchers working with large public record datasets, such as finance, lobbying, and contracts data. They require automated entity resolution and connection mapping. The GitHub repository for OpenPlanter is EliabLemus/openplanter.