Context infrastructure

Context infrastructure
for AI agents.

Tensor gives AI agents the context they need from the systems they depend on, before they reason or act.

Our first focus is legal AI agents, where every input must be accounted for before the agent acts.

Book a walkthroughSee how it works

Built with support from

Know if your agent has enough context to act.

Tensor is GammaLex's context runtime. One call, and the agent gets the evidence and a sufficient, insufficient, or incomplete answer back.

The agent asks Tensor

One call before the agent reasons or acts: who is asking, and what about.

GammaLexAgentSDK

It works out what the request requires

It works out what must be present and what the agent may use, before anything is read.

GammaLexOrchid MSA — Project LanternBefore any read

Requirements

  • Executed agreementRequired
  • Amendment chainRequired
  • Change-of-control clauseRequired
  • Disclosure scheduleRequired
  • Counsel adviceRequired

412 records permitted · 1,906 excluded

It assembles the context

It gathers only what the request depends on, from the connectors the agent may read.

GammaLexcontext6 selected

Selected context

  • Executed MSA, Calder ↔ Orchid Cloud Ltd
  • Amendment No. 3 (2026-03-18)
  • Disclosure schedule item 7.3 (2026-02-03)
  • MSA §14.2 — change of control
  • Outside counsel advice

4 of 6 shown · 1 withheld

It checks whether it's enough

It names anything missing, contradictory, stale, ambiguous or restricted — or not checkable — before the agent relies on it.

GammaLexRUN_7D41C2Checked

Requirements

  • Executed agreementEstablished
  • Amendment chainNot found
  • Change-of-control clauseContradictory
  • Disclosure scheduleStale
  • Counsel adviceRestricted

1 of 5 established

One complete run.

The same run, as the agent receives it.

GammaLexGammaLex AI

Can we rely on the Orchid MSA change-of-control position for the diligence memo?

INSUFFICIENT

Diligence agent · Orchid MSA — Project Lantern · via SDK · 6 records

To resolve

  • Provide Amendment No. 2 (Data room)
  • Confirm which governs: MSA §14.2 or Disclosure schedule 7.3
  • Counsel advice is privileged — request access or a summary from the matter lead

Every Context Run ends with one of three outcomes.

Insufficient if any required requirement isn't established; otherwise incomplete if any can't be checked; otherwise sufficient.

Sufficient

Every required requirement is established from current, permitted evidence.

Proceeds.

Incomplete

A required connector couldn't be checked, so the result can't say.

Doesn't guess; the reason says what to connect or retry.

  • Not authorized
  • Unavailable
  • Not connected
  • Out of budget

Insufficient

A required requirement isn't established.

Stops; the reasons come back with the evidence.

  • Not found
  • Contradictory
  • Stale
  • Ambiguous
  • Restricted

What every run gives an agent.

  • The relevant information

    Only the records the request depends on, from what the agent is permitted to read.

  • A clear outcome

    Sufficient, insufficient, or incomplete before the agent acts, with the reason for each requirement.

  • Enforced permissions

    What the agent isn't permitted to see never reaches it, and every exclusion is recorded.

  • A record of every run

    What was used, what was left out, and why. The same request, Context Model and snapshot replay to the same result.

Not search. Not guardrails. The step in between.

Search finds what looks relevant. Guardrails filter what comes out. GammaLex checks that the context is enough before the agent acts.

Built for the systems your agents already use.

Your agent makes one call. It reads from the systems the request depends on; it never writes to them.

SDK · API · MCP · CLI · Connectors
SDK
Python. One call returns the context and the Context Run result.pip install "gammx[client]"
API
Run a Context Request over HTTP. The same request, bundle and snapshot give the same result.POST /v1/contextPOST /v1/records · GET /v1/runs/{id}/context-run
MCP
Five tools an agent can call directly: run a Context Request, inspect a run, fetch its evidence, replay it, or sandbox a what-if.get_context · inspect_context · get_evidence · replay_context · sandbox_context
CLI
Run requests locally with the gammx CLI, or embed gammx.Tensor in your own process.gammx run --bundle <bundle> --task <task> --actor <actor> --scope <scope>

Legal records

  • CourtListener
  • PACER

Documents & teams

  • Google Drive
  • GitHub
  • Slack

Data & web

  • Web pages
  • PostgreSQL

Yours

  • Pushed records
  • Local files

Connectors available today. Each workspace connects the ones it uses; PACER is opt-in and fee-capped.

Open source · OpenContextually · In development

The same check, for coding agents. OpenContextually decides what an agent should read before it touches a codebase. No LLM, no network, deterministic — and it never edits your code.

View on GitHub

Give your agents context they can rely on.

Taking on design partners running agents in high-consequence workflows, and teams integrating the layer into their own product.

Book a walkthrough