SUTRA/Overview
Autonomous Engineering Control Plane
Let AI agents write the code. Keep humans in charge of what ships.
SUTRA gives coding agents their own identity, scoped access and an evidence trail, and blocks any merge until a human approves.
Works alongside your existing GitHub repos.
CLEAR DIVISION OF RESPONSIBILITY
What GitHub Does vs. What SUTRA Does
GitHub is where your code lives and collaborates. SUTRA is the protective control plane that governs how AI agents interact with that code safely.
What GitHub Does
The Source Code & Team Collaboration Hub
Stores Git Repositories
Hosts your main branches, commit history, releases, and repository tree.
Hosts Pull Requests & Issues
Provides issue boards, discussion threads, and team code review comments.
Runs GitHub Actions & Webhooks
Triggers basic CI runners, deployment webhooks, and status checks on push.
Manages Human Permissions
Controls team read/write access and organization memberships.
Limitation: GitHub treats AI agents like generic users with static tokens. It cannot bound agent execution, inspect task intent, or prevent silent rogue changes.
What SUTRA Does
The Safe Autonomous Engineering Control Plane
Zero Direct Access to Main Branches
Agents are constrained to ephemeral, isolated workspaces with short-lived credentials.
Structured Codebase Context
Agents work within the context of the actual repository — real files, real branches, and real history — rather than hallucinating structure from scratch.
Verified Commit Provenance
Every commit is linked to the agent session that created it — recording which agent made the change, when, and what CI checks ran.
Mandatory Human Approval Gate
Agents are programmatically blocked from self-merging. A verified human must review and approve.
SUTRA Advantage: Works alongside your existing GitHub repositories without replacing your source code host.
Writing code is only part of engineering.
AI systems can generate code quickly. Real software engineering also requires identity, authorization, repository access, version control, testing, review, auditability, and a controlled path to completion.
SUTRA focuses on the layer between an AI model and a real engineering workflow.
CORE CAPABILITIES
Give AI agents the ability to do real engineering work.
Controlled Agent Identity
Every external agent gets a distinct SUTRA identity and short-lived session instead of operating through a human's credentials.
Identity GuideRepository-Scoped Authority
Agents only receive access to repositories and capabilities explicitly granted to them.
Authority GuideReal Engineering Workflow
Agents can work through branches, commits, Changes, Pull Requests, CI, review, and merge.
Workflow GuideHuman Control
Agents cannot approve their own work or bypass repository, review, or merge controls.
Governance GuideSTEP-BY-STEP LIFECYCLE
How SUTRA works
01
Connect your Agent
Register an external AI agent with SUTRA.
02
Give it bounded access
Choose the repository and capabilities that agent is allowed to use.
03
Let it work
The agent receives a short-lived session and performs real engineering work.
04
Verify the result
SUTRA records the commit, changed files, diff evidence, CI state, and review state.
05
Keep the human in control
A human reviews and approves the work before protected changes are merged.
RELIABILITY
When agents go wrong, SUTRA is what stops the damage.
Autonomous agents make mistakes. These are the four patterns we see most often, and how SUTRA constrains them.
Runaway loops and retries
Sessions are short-lived and bounded. An agent stuck in a retry loop has its session revoked after 15 minutes (absolute TTL) or 120 seconds of inactivity — it cannot run forever or accumulate unbounded cost.
'Success' that isn't
An agent saying it finished is not proof. SUTRA records commits, diffs, and CI results independently of what the agent reports — so you can see the actual outcome, not just the agent's claim.
Wrong turns
Access is scoped to specific repositories and capabilities granted at registration. A bad decision cannot reach beyond its explicit boundary — other repos and capabilities remain inaccessible.
Recovery
Agent access can be revoked and sessions expire automatically. Every action leaves an immutable audit trail, so you can always see exactly what happened and when.
See how a failed CI run is handled in the live agent trail →
Power without uncontrolled access.
Agents should be capable. They should also be bounded.
Identity→Authority→Execution→Evidence→Verification→Human Control
IDENTITY
An external agent is represented by a distinct SUTRA Agent identity.
AUTHORITY
Access is explicitly scoped to repositories and capabilities.
EXECUTION
Agents can perform real engineering operations through controlled interfaces.
EVIDENCE
Engineering work is tied to real commits, changed files, diff statistics, CI, and review state.
VERIFICATION
System policies and human review determine whether work can progress.
CONTROL
Agent access can be revoked and sessions expire.
Every change leaves evidence.
SUTRA connects an agent's engineering action to the artifacts and checks surrounding that action.
Agent
Registered external agent with scoped capability↓
Commit
Real Git commit↓
Changed Files
File-level change evidence↓
Diff
Actual additions and deletions↓
CI
CI results↓
Review
Human review↓
Merge
Verified merge stateHumans remain the final authority.
Agents can perform engineering work, but they do not become the final approver of their own work.
Repository authorization, policy evaluation, review, and merge controls remain separate from agent execution.
THE TEAM
Built in the open, by people who use it.
SUTRA is used by its own team to build SUTRA. Every design decision comes from real agent governance experience.
K
Kartikay Achint.
Founding Engineer, SUTRA. Building the engineering control plane for the autonomous coding era.
FREQUENTLY ASKED QUESTIONS
Questions & Answers
SUTRA is owned and developed by AchintAI as the authoritative control plane for autonomous software engineering.
GitHub serves as the repository substrate for code and branches. SUTRA is the control plane governing agent permissions, AgentSession leases, cryptographic change provenance, and merge approval policies.
Build with AI. Keep control.
SUTRA gives your team the safe, governed execution environment needed to let autonomous agents build and ship real software.
Already have an account? Sign in