Feed it a question. Get back scored claims, entity graphs, and a synthesized report — using your own LLM.
Research
pip install behive==0.6.0
Scored, source-linked statements
Names and relationships
Model-generated synthesis
Persisted research missions
The Problem — schematic comparison
Unstructured answer
"Based on my training data, NVIDIA's GPU market shows promising growth with several key factors driving demand..."
Schematic BeHive artifacts
[claim] Example claim → original source URL
[entity] Example entity → linked research mission
[link] Example relationship → supporting claim
How It Works
The planner breaks a topic into research tasks. Search backends and topic-specific API adapters discover candidate sources.
Fetch candidate pages and documents with configured HTTP or optional browser backends. Some sources need credentials or extra dependencies.
Your configured model extracts claims from source content. Quality heuristics help rank extracted statements for review.
Similarity checks reduce repeated claims. Deduplication is heuristic: similar statements can still need human review.
Your configured model synthesizes extracted claims into a report. Check its conclusions against the linked original sources.
Persist claims, entities, and relationships. The PostgreSQL-backed API exposes entity and network lookups across missions.
Capabilities
Available interfaces in v0.6.0. Results depend on source access, configuration, and your model.
| Artifact | What you can do |
|---|---|
| Claims | Search saved claims and inspect their scores and sources. |
| Mission status | Poll progress or subscribe to API server-sent events. |
| Report | Retrieve the synthesized report for a completed mission. |
| Entity network | Explore stored entities and relationships through the API. |
| Mission history | List saved missions and reuse the knowledge store. |
A quality score is a heuristic, not a probability of truth.
Check original sources. Extraction and synthesis can be incomplete or incorrect.
Integrations
CLI
Run missions with depth, scale, and local or remote storage options.
REST API
Start research, poll status, and retrieve reports or stored claims.
Server-sent events
Subscribe to the API mission events endpoint.
JSON-RPC over HTTP
Connect to the /mcp endpoint using client-specific configuration. Compatibility must be checked with your client.
BeHiveClient
Use the packaged client for local or remote research.
Entity + network API
Query the PostgreSQL-backed knowledge store across missions.
MCP endpoint (JSON-RPC over HTTP)
Start the service below, then connect to http://127.0.0.1:8090/mcp. Configuration is client-specific; this is not a universal desktop-client config.
curl http://127.0.0.1:8090/mcp -H 'Content-Type: application/json' -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Tools: research_topic, search_knowledge, mission_status, get_report, list_missions. API mission progress uses SSE; the MCP bridge responds with JSON.
Current bridge limitation: it does not forward API credentials, so research_topic cannot start an authenticated API mission when BEHIVE_API_KEY is enabled. Use the authenticated REST API for that setup.
Self-Hosting
Python 3.10+ in a virtual environment. Configure your model and provider credentials. Local CLI research can select DuckDB; API + MCP requires PostgreSQL.
python -m venv .venv
source .venv/bin/activate
pip install behive==0.6.0
export OPENAI_API_KEY="YOUR_PROVIDER_KEY"
behive config --quick
behive research "your research question" --depth 3 --no-dbPOSIX shell example. Use credentials for your selected model; OPENAI_API_KEY is one provider example. Optional source-processing extras: pip install 'behive[harvest,process]==0.6.0'.
Provision PostgreSQL and create a database and role first. Replace the placeholder connection string with your credentials.
export BEHIVE_DB_URL="postgresql://USER:PASSWORD@127.0.0.1:5432/behive"
behive db init
behive serve --host 127.0.0.1 --port 8091 --mcp-port 8090API: http://127.0.0.1:8091 · Docs: /docs · MCP: http://127.0.0.1:8090/mcp. Keep local by default; configure authentication and network controls before remote exposure.
curl http://127.0.0.1:8091/research -H 'Content-Type: application/json' -d '{"query":"your research question","depth": 3}'Use the returned mission ID with /research/{mission_id}/status, /research/{mission_id}/events, and /research/{mission_id}/report. When BEHIVE_API_KEY is configured, include its Bearer token on research requests. The MCP bridge does not enforce this API key; keep its endpoint private.
Requirements
Python 3.10+, a configured LiteLLM-compatible model, and accessible sources. The --no-db option selects local DuckDB storage in ~/.behive/research.duckdb.
Configure PostgreSQL and initialize its schema before starting the service. MIT-licensed software; model, search-provider, hosting, and network costs are yours. Optional extraction and browser dependencies may be needed for specific sources.
Configure a model. Research with sources. Review the results.
pip install behive==0.6.0