Features / Research
Research workers,
multiplied.
Universal autonomous workers can plan research, delegate configured tools, fetch and cross-reference evidence, persist memory and pause at governance gates before producing a cited, reviewable answer.
Capabilities
How Fathom researches the web
Search services such as Linkup, Parallel.ai, Exa, Tavily or Serper can be enabled per deployment. Workers combine available results and retain source provenance.
spawn_agent lets a coordinator delegate subtasks within configured depth and budget; background work can report through enabled notifications.
web_fetch, web_crawl and web_feed retrieve pages, sitemaps or RSS when the corresponding network access and connectors are configured.
parse_html and extract_json turn fetched documents into structured evidence for worker reasoning and later review.
A configured Chrome CDP session can handle interactive pages; analyze_image can inspect screenshots, charts and documents.
code_symbols and repo_map let workers combine repository context with web sources and document extraction when those tools are available.
Comparison
A worker workflow vs. manual work
Manual
- Hours per topic, one tab at a time
- Single search engine, first-page bias
- Copy-paste into spreadsheets
- No verification, stale data
- Nothing remembered next week
Fathom
- Delegated subtasks within configured budgets
- Available search tools with source provenance
- Structured findings, memory and exports
- Evidence gaps and reviewable confidence
- Persisted context carries knowledge forward
Under the hood
How a worker plans and delegates
Plan
LLM decomposes the query into sub-tasks; subtasks persist in SQLite.
Fan out
Coordinator spawns sub-agents (JoinSet or separate OS processes).
Collect
Budget-capped summaries flow back; context stays small.
Reflect
Goal Mode: an LLM judge checks the result against the goal and runs gap-filling rounds.
Synthesize
Findings merge into summary.md + findings/, absorbed into memory.
Agent roles
Composable roles on one runtime
Every agent runs a role-specific prompt on top of the same runtime. Roles can be overridden per deployment via [agent.role_models].
Decomposes the request into scoped subtasks, delegates within configured budgets and synthesizes a reviewable answer.
Runs web research, OSINT or lead-discovery workflows with source citation as configured examples.
Cross-references findings, spots patterns and contradictions, assigns HIGH / MEDIUM / LOW confidence labels via the system prompt.
Adversarial fact-checking: Verified / Partial / Unverified.
Structures the report: Summary / Findings / Analysis / Sources / Gaps.
Stable, context and volatile prompt layers can keep role instructions separate from task state and runtime context.
Use cases
Research as one worker use case
Market & competitor scans
Ask a worker to track pricing, positioning or hiring, then compare persisted findings between runs.
Due diligence
Assemble company facts, news, filings and key people into a cited dossier with explicit gaps for review.
Lead research
Identify decision-maker evidence at target companies; enrichment, verification and outreach remain configured workflows subject to approval.
Tech scouting
Map repositories, extract symbols and summarize documentation with code_symbols and repo_map when available.