TC Agent Coding Workbench Caller-bound coding control plane
Control plane online

An agent coding workbench built for controlled execution.

Commands, edits, search, diffs, verification, artifacts, and recovery guidance stay inside a caller-bound workflow with proof on every move.

Open the workbench Show the gates
20+MCP surfaces
G1-G6Proof loop
RustNative helpers
LiveHTTP runtime
shell_runverify_buildbeadsguarded_editdiff_viewtags_updatecodebase_searchproject_queryfile_viewmanual shell_runverify_buildbeadsguarded_editdiff_viewtags_updatecodebase_searchproject_queryfile_viewmanual
Product position

The control plane for AI engineering work

The grounded evidence engine supplies proof. The agent coding workbench turns local implementation into a bounded loop: inspect, edit, verify, recover, and report.

Launch pillars

Bounded executionEvery operation starts inside the selected repo and stays there.
Visible progressTasks, proof, and handoff state do not disappear into chat scrollback.
Proof gatesPlan, act, verify, commit, report. The rhythm is already built.
Playbook routingReviews, bug hunts, wiki work, contracts, and SoT flows land in the right lane.

Why teams notice

Feels premiumStructured surfaces, bounded output, clean envelopes, and clear state ownership.
Looks operationalWorkspace standards, task state, gates, and service proof make it feel production-ready.
Built with musclePython drives the service layer. Rust handles the native helper core.
Ready to hand offArtifacts, structured responses, and proof logs make the work portable.
Operator console

Everything the agent needs is already on the bench

Search, edit, shell, diff, build, memory, docs, artifacts, tags, cleanup. One cockpit. One boundary model. One serious workflow.

Inspect lane

Repo visibility with command-center polish.

Fast discovery. Clean readback. No messy raw dumps.

Search codebase_search Find the right file, symbol, or evidence path fast.
Read file_view Exact repo readback without noise.
Read code_view Code slices with symbol-level focus.
Docs doc_view Docs surfaced like a navigable product.
Diff diff_view Git changes, shaped for review.
Scan project_query Project scans and artifacts on demand.
Launch sequence

Proof is part of the product

The workflow has rhythm: scope, act, verify, iterate, commit, report. It feels fast because the guardrails are already there.

G1

Plan

Pick the mission. Lock scope. Name the win.

Bead required
G2

Action

Move with intent. Keep the surface tight.

Changes named
G3

Verify

Run the checks. Keep the proof.

Evidence first
G4

Iterate

Tune, rerun, and stay inside the lane.

Bounded loops
G5

Commit Gate

Refresh the map. Make the verified state durable.

tags_update
G6

Report

Close with evidence. Leave the next operator ready.

Close cleanly
Boundary model

Power, boxed and ready

The product promise is simple: give agents serious coding power without letting the workspace become untraceable.

Caller-owned state under cwd
.beads/ .cache/agent/ artifacts logs Tags/ SCRATCH.md
Capability surface MCP runtime
tool registry structured envelopes schema contracts bounded errors redaction
Native helper core Rust-first
guarded_edit diff_view workspace_patch verify_build tags_update
Durable guidance docs and memory
AGENTS.md SourceOfTruth/ KnowledgePack Long-Term Memory playbooks
Grounded Evidence

Proof engine

Grounded knowledge, retrieval, durable wiki memory, and corpus proof.

+
Agent Coding Workbench

Execution control plane

Bounded tools, proof gates, task memory, verification, contracts, and service operation.

+
Operating Playbooks

Repeatable expert workflows

Trigger routing, repeatable workflows, helper review, wiki curation, and portable agent behavior.

Launch-ready system

The coding workbench that makes agents production-ready.

Raw agent capability becomes a polished engineering control plane: fast to use, hard to misuse, easy to verify, and impressive on sight.