Thesis gives you four ways to retrieve and synthesize knowledge: unified search across your graph and indexed sources, Oracle research jobs for longer synthesis tasks, Tracer for GitHub-oriented code search, and Deep Research for Exa-backed paper and web discovery. All four are accessible from the Knowledge panel in the web app and as MCP tools. This guide shows you how to use each one.

​
The Knowledge panel

Open the Knowledge panel from any node view by clicking the search icon in the node sidebar. It has four tabs:
TabWhat it does
AskUnified search, queries your graph and all indexed sources together
OracleLong-running research synthesis jobs with three depth modes
TracerGitHub code search across indexed repositories
DocumentQ&A against a single indexed source
1

Run a unified search

Unified search fans out across two knowledge branches simultaneously:
  • Graph-as-KB, committed Thesis nodes and their content, embedded into searchable chunks.
  • Nia proxy, external papers, repositories, docs, and datasets you’ve indexed.
Results are deduplicated, scores merged, and returned in a consistent shape regardless of source.Call thesis_search directly:
{
  "name": "thesis_search",
  "arguments": {
    "query": "learning rate warmup effect on GRPO reward stability",
    "node_id": "<optional-node-id-to-scope-search>"
  }
}
The result includes ranked hits with source IDs, titles, excerpt snippets, and relevance scores. Each hit links back to either a graph node or an indexed source so you can read the full content.In the Knowledge panel, type your query in the Ask tab and press Enter. Results appear inline with source attribution.
2

Start an Oracle job for deeper synthesis

Oracle jobs run a research pipeline against your knowledge base and return a synthesized report. Choose the depth that matches your task:
ModeBest for
quickFast summaries of a narrow question
deepMulti-source synthesis with more iterations
oracleMaximum depth, slower, most thorough
Start a job:
{
  "name": "thesis_oracle_start_job",
  "arguments": {
    "query": "Compare cosine and linear LR schedules for GRPO training across recent papers",
    "mode": "deep",
    "node_id": "<node-id-to-attach-results>"
  }
}
Check the job status with thesis_oracle_job_status, or stream live updates with thesis_oracle_job_stream. In the Oracle tab, you can see job history and results without leaving the node view.
Generated Oracle reports land as files and assets on the associated node. You can attach them to other nodes or include them as evidence when writing up a synthesis insight node.
3

Search code with Tracer

Tracer runs GitHub code search across your indexed repositories and returns a structured brief, useful when you want to find how a concept is implemented across codebases.
{
  "name": "thesis_tracer_search",
  "arguments": {
    "query": "cosine annealing scheduler GRPO",
    "repos_csv": "huggingface/trl,openai/spinningup"
  }
}
Retrieve the result when it’s ready:
{
  "name": "thesis_tracer_status",
  "arguments": {
    "job_id": "<tracer-job-id>"
  }
}
In the Tracer tab, enter a query and the repos you want to search. The brief surfaces matching file paths, line-level excerpts, and a synthesized summary of how the concept appears across the codebase.
Tracer requires a GitHub token connected in Settings → API Keys. If Tracer returns no results for a repo you’ve indexed, verify that the connected GitHub account has access to that repository.
4

Run Deep Research for paper and web discovery

Deep Research combines Exa-powered search with full-text source fetching and Anthropic synthesis. It’s a deterministic pipeline, not another ReAct loop, so results are consistent and reproducible.Use it when you want to discover external papers and web sources on a topic and receive a synthesized report without indexing sources manually first.From the agent chat, ask: “Run a deep research job on cosine LR schedules in reinforcement learning from human feedback.” The agent calls deep_research_start internally, which:
  1. Runs Exa search to discover relevant papers and web sources.
  2. Fetches full-text content via Exa.
  3. Synthesizes a report using Anthropic.
  4. Writes the report to research_context/deep-research-<slug>.md in the project filesystem.
The report lands as a registered asset and appears as an asset_card in chat. You can attach it to any node in your graph.
Deep Research requires both Exa and Anthropic to be connected in Settings → API Keys. If either is missing, the job returns an error before fetching any sources.
5

Attach reports to nodes

Any report generated by Oracle, Tracer, or Deep Research can be attached to a node as an artifact. In the node editor, open the Artifacts panel and link the file. This creates a durable connection between the evidence and the claim or experiment it supports, so future searches and sub-agents can trace where a finding came from.

​
Choosing the right tool

ToolSpeedDepthBest when
thesis_searchFastShallowYou know roughly what you’re looking for
thesis_oracle_start_job (quick)MediumMediumYou want a synthesis in under a minute
thesis_oracle_start_job (deep / oracle)SlowDeepYou want a thorough literature synthesis
thesis_tracer_searchMediumCode-focusedYou need to find implementation patterns across repos
Deep ResearchSlowExternalYou want to discover new papers and web sources