Sovereign AI software engineering

Plan, code, host, and review with autonomous agents.

TensorPlane unifies the complete AI development lifecycle on infrastructure you control. From MCP-native task planning to autonomous coding agents in isolated microVMs, zero-markup model routing, and automated PR reviews — EU-hosted or on-premises.

Start 14-day trial See pricing

14-day trial · No credit card required · Connects to GitHub in minutes · Zero token markup

$ tensorcheck --pr 4821
checkout              sandboxed · 38 files changed
policy check          passed (14 rules)
findings              3 (0 blocking)
▸ session.ts:88       token TTL bounded to MAX_TENANT_TTL
▸ retry.ts:41         retry exponential backoff verified
✓ posted to GitHub Check Run

- const sessionTtl = req.requestedTtl ?? 86400;

+ const sessionTtl = Math.min(req.requestedTtl ?? 3600, MAX_TENANT_TTL);

GUEST ISOLATIONDedicated Kernel 6.1Hardware-enforced tenant boundary

NETWORK CONTAINMENTeBPF Isolated TAPEgress strictly locked to TensorGate

BOOT LATENCY~150ms from warm poolZero cold-start container lag

HOST INTEGRITYZero Guest CredentialsUntrusted code cannot touch host keys

[tb-8842] 00.002s spawn Firecracker jailer with dedicated netns
[tb-8842] 00.148s guest kernel booted, rootfs mounted ro
[tb-8842] 00.152s eBPF ingress/egress filter attached
[tb-8842] 00.160s agent execution started in disposable overlay

ACTIVE MODEL ROUTEgpt-5-codexOpenAI Responses API (EU Resident)

CREDENTIAL CUSTODYVaulted in OpenBaoNever exposed to sandboxes or devs

TOKEN ATTRIBUTION3,420 tokensPR #4821 · Attributed to Team Backend

COST FOR PR #4821€0.042Exact provider cost · TensorPlane fee: €0

$ tensorgate status --org acme
providers             3 (EU endpoints) · keys vaulted
routing               simple → small · complex → frontier
cache                 org-wide · prompt + context
spend                 €41,806 MTD · 0 limit breaches
✓ calls metered and attributed

LINEAGE TRACERequirement → Diff → PRWalkable audit trail from commit to spec

ADVERSARIAL INTEGRITYAuthor Fencing ActiveAuthor cannot approve their own pull request

WORKTREE SAFETYZero Data LossOrphaned worktrees quarantined atomically

MODEL COST ATTRIBUTION€4.12 Task TotalEvery model cent tied directly to the ticket

# mcp__tensorline__task_show { "id": "TP-102" }
task                  E2E quarantine worktree suite
plan                  accepted · 4 sub-tasks
runs                  2 (agent) · microVM tb-8839
review                1 blocking finding → fixed
✓ merged 9a4f21 · total model cost €4.12

The AI engineering lifecycle

From spec to merged pull request — governed at every step.

Autonomous engineering requires a closed loop. TensorPlane connects planning, coding, hosting, and reviewing into a single continuous system with zero model markup and cryptographic auditability.

01 / TensorLine

Plan & Scope

Work tracking for agents — not dead Markdown files.

Specifications and tasks live as database-backed, audit-logged entities rather than Markdown files scattered in repositories. Agents and engineering leads coordinate work dynamically through 20 native MCP tools.

20 native Model Context Protocol (MCP) tools for agent planning and task claiming
Adversarial review constraints: authors cannot sign off on their own pull requests
Worktree lifecycle tracking with atomic zero-data-loss quarantine

Explore TensorLine

task                  TP-104 · Implement rate-limiter & token TTL
plan                  accepted · 4 scoped sub-tasks
fencing_token         fence_84920b1 · author-bound
runs                  2 (agent runs) · microVM #tb-8842
worktree              quarantine-safe · zero data loss

02 / TensorCode & TensorBoot

Implement & Execute

Autonomous coding agents running in isolated microVMs.

TensorCode scopes implementation from TensorLine tasks and executes coding agents in dedicated TensorBoot microVMs — kernel-level isolation booted in ~150ms on your own infrastructure or EU cloud.

Hardware-isolated microVMs with dedicated guest kernels (Firecracker & Cloud Hypervisor)
~150ms instant boot from warm snapshot pools with zero cold-start latency
eBPF per-tenant network namespaces: agents can reach only authorized endpoints

Explore TensorCode

sandbox               tb-8842 · dedicated kernel 6.1.102
boot_latency          148ms from warm pool
isolation             microVM · per-tenant eBPF network TAP
egress                locked to TensorGate relay only
credentials           none in guest · zero host exposure

03 / TensorOrigin

Store & Version

Sovereign Git storage on immutable S3-compatible object stores.

Source code, branches, and commits are persisted in an immutable object-store Git daemon. Repositories carry cryptographic provenance and maintain disposable local caches for high-throughput agent runs.

Object storage authoritative: immutable WAL and pack files with conditional index publication
Disposable local Git caches: complete cache loss recovers safely from object storage
Zero dependency on external foreign hosts for air-gapped and sovereign operations

Read TensorOrigin docs

repository            repo_9f82ab · git v2 smart-HTTP over TLS
storage_engine        S3-compatible immutable pack/WAL (OVH Gravelines)
durability            atomically conditional index write
local_cache           disposable · recovers acknowledged writes
auth                  capability tokens bound to (org_id, repo_id)

04 / TensorCheck

Verify & Gate

Adversarial PR code review in full repository context.

Before any code merges, TensorCheck checks out the PR into a sandbox, inspects the whole repository, and posts deduplicated findings alongside a blocking GitHub Check Run.

Full repository context: evaluates whole files and dependency graphs, not just isolated diff hunks
Incremental re-reviews: only inspects changes since the previous pass, keeping comment noise near zero
Severity-gated GitHub Check Runs: prevent merges until critical findings are resolved

Explore TensorCheck

checkout              sandboxed · 38 files changed
policy check          passed (14 security rules)
findings              3 (0 blocking, 1 security advisory)
▸ session.ts:88       token TTL bounded to MAX_TENANT_TTL
✓ posted to GitHub Check Run (conclusion: success)

05 / TensorGate

Route & Protect

Centralized LLM gateway with zero token markup.

Every prompt across the lifecycle travels through your own vaulted keys with zero inference markup. TensorGate routes simple tasks to cheaper models, enforces spend caps, and attributes every cent to tasks.

Zero token markup (BYOK): you pay model providers directly at exact cost
OpenBao key custody: credentials are vaulted centrally and never exposed to sandboxes or laptops
Graceful spend limits: soft budgets alert and route economically without killing active jobs

Explore TensorGate

providers             OpenAI, xAI, Moonshot, Z.AI (EU resident)
routing               simple tasks → economical · complex → frontier
token_markup          €0.00 (0% surcharge · direct provider billing)
attribution           task TP-104 · PR #4821 · Team Backend
caching               org-wide prompt & context caching active

The platform suite

Five components. One product, one price.

Everything below is included in a single platform subscription — €35 per contributing developer per month. No per-component SKUs, no token markup — with policy, entitlements, audit, and spend attribution built into the control plane underneath.

TensorCheck

Code review

Reviews pull requests in full repository context and posts findings — not noise — so humans start from findings, not from scratch.

Explore TensorCheck

TensorBoot

Coming soon

Boots each agent run into its own microVM — kernel-level isolation in milliseconds, on servers you control in the self-hosted deployment.

Explore TensorBoot

TensorGate

LLM gateway

Routes every model call through your own keys at cost — zero markup — and sends simple tasks to cheaper models.

Explore TensorGate

TensorLine

Coming soon

Turns agent plans, tasks, and reviews into a live linked tracker instead of Markdown files nobody reads.

Explore TensorLine

TensorCode

Coming soon

Our coding agent harness: runs AI agents against your repositories under your policies, with every action logged and attributable.

Explore TensorCode

Enterprise governance

Engineered for regulated teams that cannot compromise on security.

Deploying coding agents inside an enterprise requires more than a prompt. TensorPlane provides the governance, auditing, and repository readiness that security and compliance teams require.

AI Readiness & Repo InitScore: 94/100

Automated repository initialization for coding agents.

Legacy codebases lack the context, test boundaries, and guidelines that autonomous agents need to succeed. TensorPlane scans your repositories for readiness signals and automatically issues bootstrap pull requests.

+ ADDAGENTS.mdRepository structure, build rules & test anchors

+ ADDCONTEXT.mdDomain glossary, system boundaries & architectural invariants

+ ADD.tensorplane/policy.tomlSandbox constraints, excluded paths & spend limits

Skills RegistryPinned Hashes

Governed Skills & MCP Registry.

Control exactly what tools agents can invoke. Maintain an approved catalog of MCP servers and agent skills with publisher RBAC, release channels (stable, canary, pinned), and cryptographic SHA-256 verification.

registry: mcp.tensorplane.internal
mcp_servers:
  - id: git-safe
    sha256: 8f4b2a9e...
    channel: stable
    rbac: team-platform

Enterprise SIEMHash-Chained

Cryptographic audit & SIEM streaming.

Every prompt, diff, tool invocation, and policy decision writes to an append-only, tamper-evident hash chain (tp/audit/v1) with external witness receipts. Real-time streaming in CEF, LEEF, and OpenTelemetry.

Splunk (CEF)QRadar (LEEF)Datadog (OTel)Sentinel

Policy GuardrailsZero Leakage

Active guardrails & secret redaction.

Versioned guardrail packs and automated entropy scanners sanitize API credentials, tokens, PII, and internal hostnames in memory before inference requests leave the perimeter.

In-memory entropy secret scanner
Closed route tables preventing SSRF
Instant sub-second fleet kill-switch

SovereigntyGDPR First

European sovereignty by design.

Hosted in the EU (OVH Gravelines) under strict data processing agreements from day one. In self-hosted mode, code and prompts never transit our infrastructure.

DPA on every customer tier
Direct to provider EU endpoints
Offline grace window / air-gap ready

Transparent pricing

Seats for people. Usage for agents. Zero token markup.

Every tool in the suite is included at €35 per contributing developer per month. Background cloud agents run inside dedicated microVMs billed by the hour — and completely free when self-hosted on your own hardware.

Platform Subscription

All Tools Included

The complete suite on your infrastructure or EU-hosted. No minimum seats — scales smoothly from your first developer.

€35 per contributing dev / month

Billed monthly via Stripe · Volume discounts to €25 · 14-day free trial

TensorCheck PR reviews on unlimited repos
TensorGate LLM gateway with zero token markup
TensorLine AI work tracking with 20 MCP tools
TensorOrigin sovereign Git storage engine
Enterprise governance, AI readiness & SIEM export
Baseline review sandbox capacity included

Start 14-day trial

Cloud Agent Compute

Pay as you go

Dedicated TensorBoot microVM sandboxes for autonomous background agents running via TensorCode.

Standard MicroVM2 vCPU · 4 GB RAM · Dedicated kernel

€0.12 / hr

Performance MicroVM4 vCPU · 8 GB RAM · Dedicated kernel

€0.24 / hr

Self-Hosted ExecutionRuns on your own servers via TensorBoot

€0.00 / hr

Metered to the second. Zero markup on model tokens — you pay providers directly via your vaulted keys.

View full pricing details

Start building with sovereign coding agents today.

Connect your repositories in minutes with our 14-day free trial. No credit card required, full suite access, zero token markup.

Start 14-day trial Talk to engineering